s3jL7gGj

1[00:01:03.504] exa65536: Rybka'll be the first SUFI participant to drop out entirely - but it never actually won one, and it'll be a long time yet before H demotes all the way for the first SUFI winner to drop out
2[00:02:18.773] K_recreant: @dark - 433 Not past random babble iirc
3[00:03:41.789] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
4[00:05:41.337] Aloril42: @exa65536 Well, Fritz might not drop if it updates..
5[00:06:15.262] Aloril42: So Rybka kind of never really drops, just is at some point too weak.
6[00:11:32.145] atanh: !cdb
7[00:11:33.540] chessdbcn (38w) • Rf1 43 ! (01-01) • Na4 37 ! (01-01) • Ra6 33 * (12-04) • Ra2 31 * (01-01) • Ra4 31 * (06-03) • Qb1 26 * (01-01) • pv Rf1 f5 exf5 exf5 • <ChessDBCN>
8[00:23:43.061] lmabacus: These two are just not interested in exchanging pieces.
9[00:26:44.801] mdt_sf: !-
10[00:26:45.003] Nightbot: NEG-A-TIVE
11[00:31:28.332] atanh: @darkranger433 Right. You did 'zero', so... Point is simply that starting from zero takes a long time. One of the longest time things is generating training games from the existing (untrained) net. If it's playing in a 30 min game and uses 25 min just to generate games, how many games could it generate? Say 100. But each would only be 0.25 min long, so no help for the 5 min game play remaining.
12[00:33:50.751] atanh: @darkranger433 Re: Training on TCEC HW for 1h. With current methods, barely better than random play. How much better? Not sure.
13[00:35:13.400] atanh: Could Scorpio actually be shifting this towards a win? First win in DivP?
14[00:36:39.614] Aloril42: It takes years for humans to train to WCC level.
15[00:36:56.057] atanh: @Aloril42 So, when Fritz drops to QL, it will be interesting to see how long it lasts there, too. A preview of Houdini's future pre-retirement stint in QL. I bet Fritz lasts a while there, too.
16[00:37:57.507] darkranger433: @atanh That makes sense, it would need to be obscene hardware the games were being played on, at LTC, just to generate a half decent net, in all likelihood
17[00:42:34.339] Aloril42: Would be nice if some day algorithms were enough good to train net further for current position and hardware enough fast.
18[00:45:07.259] atanh: AlphaZero took, IIRC, about 1 hour (or was it 4 hours?) to 'learn chess', which I think meant it was playing about equal to SF 8. (Could be wrong on the details, but the general point will stand.) Then to be able to dominate SF 8 took something like a few days more. Someth like that. And that was with a bajillion super-duper brand new TPUs. So, 'zero' on 2 TCEC GPUs would take a long long time.
19[00:46:01.706] atanh: But also depends on net size, so if you tried a very small net, who knows, maybe you could get some rinky dink NN that could beat the average human with no chess experience in a short time. Maybe not 1hr though. Dunno.
20[00:46:06.953] haxboi25: is alpha zero a hoax?
21[00:46:32.356] atanh: Is Leela a hoax?
22[00:46:53.880] haxboi25: no
23[00:50:48.489] atanh: Leela (Lc0) wouldn't exist if Alpha Zero hadn't published their results (how to make an engine like A0) first. And AlphaZero is actually derived from its predecessors AlphaGo Zero and AlphaGo. And AlphaGo beat the best humans decisively in Go a couple (a few?) years ago. So, no, A0 is not a hoax. Google just has mega $ and super-fancy hardware.
24[00:51:18.622] Aloril42: @atanh 4h to SF8, 9h whole training IIRC.
25[00:51:48.585] Aloril42: So they could play CC and train new net from scratch for each position in the game..
26[00:51:54.028] AngraMainyu2018: !boom5
27[00:51:54.289] Nightbot: diamondFire diamondFire diamondFire diamondFire diamondFire bueRocket
28[00:51:57.765] Aloril42: tcecB
29[00:52:02.651] enzodivasto: !gohou
30[00:52:02.893] Nightbot: erckHypers
31[00:52:24.714] atanh: Ok, thx for correction, @Aloril42. Are you sure about 9h for whole training? Hmm, maybe I'm mixing up the A0 chess training time with the A0 Go training time.. Seems likely.
32[00:52:32.766] enzodivasto: !boom7
33[00:52:32.953] Nightbot: tcecBL tcecB tcecW tcecB tcecBL
34[00:52:39.052] d_wulff: !time
35[00:52:40.379] eval_bot 80/168 games left • Average duration: 02:59:39 • Estimated division end: Tue, 24 Sep 2019 02:51:47 GMT • Shortest game: 00:59:55 • Longest game: 04:06:54
36[00:53:01.994] Aloril42: A0 Go was longer IIRC.
37[00:54:19.927] atanh: Houdini?!?! Pulls a rabbit out of a hat to distract Scorpio while he dunks him in a giant milk jug full of water, locked in a straitjacket!
38[00:55:06.749] atanh: Ok, thx. 4 and 9, I can remember that. :)
39[00:55:45.357] haxboi25: DANG!
40[00:56:33.042] haxboi25: houdini had a trick up his sleeve
41[00:57:01.636] xAbu3zzh: PogChamp
42[00:57:05.559] lmabacus: Just knew that leaving that many pieces still on the board was going to end poorly for Scorp
43[00:57:38.233] atanh: SF near 150. Maybe mate score soon. OOP M#86
44[00:57:48.476] haxboi25: !eval
45[00:57:50.499] Aloril42: What are you talking about? Black has more material! ;P tcecK
46[00:57:55.351] chessdbcn (66b) 2.99 d11 (tb 0) pv Rxe6(e7e6) Rxe6 Qxe6+(g6e6) <Lc0_v0.21.3 (J13B.2-136)>
47[00:58:03.074] eval_bot (66b) 62.12 d35 (tb 226342) pv R7xe6(e7e6) Qxe6+(g6e6) <SF_190826>
48[00:58:40.182] Aloril42: Oh, now slightly ahead in material ;)
49[00:58:47.290] atanh: M#30
50[00:59:21.114] Aloril42: NomNom FightCookie
51[00:59:37.728] atanh: PogChamp FightCorn
52[00:59:41.933] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
53[00:59:42.905] haxboi25: please dont end this i wanna see some checkmates
54[01:00:04.260] AngraMainyu2018: There will be no checkmate, I'm afraid
55[01:00:05.448] Aloril42: ScorpioNN is probably going to resign.
56[01:00:18.490] haxboi25: engines can resign?
57[01:00:33.105] enzodivasto: how resign?
58[01:00:34.709] atanh: Only xboard engines can, but yes.
59[01:00:35.014] AngraMainyu2018: Scorpio can
60[01:00:35.119] Aloril42: Xboard engines can in theoryt.
61[01:00:38.656] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
62[01:00:50.426] haxboi25: neat
63[01:01:04.486] enzodivasto: what is xboard?
64[01:01:17.285] Aloril42: Probably no SyzygyTB adjudication unless something is exchanged.
65[01:01:24.275] atanh: a protocol for controlling engines, like UCI.
66[01:02:02.444] Aloril42: Now can be SyzygyTB adjudication.
67[01:02:06.672] lmabacus: Oh, it's going to be TB adjudication
68[01:02:13.492] Aloril42: Heh.
69[01:02:15.473] alchemist888: Scorpio style: draw, draw, draw... blunder and lose
70[01:02:21.032] atanh: !uci
71[01:02:41.424] Aloril42: (script) 89: 5. Houdini 3800(+1) - 8. ScorpioNN 3715(-2) tcecW - tcecRIP https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-89-houdini-6-03-scorpionn-v3-0-1-n_maddex_int8/
72[01:02:51.669] atanh: Ah, only subscribers can !uci
73[01:03:14.379] tinduz_r: How many games in total does every engine play?
74[01:03:22.348] atanh: And there's no !xboard, btw
75[01:03:28.089] Aloril42: !uci
76[01:03:28.437] Nightbot: The UCI protocol as published by Stefan-Meyer Kahlen April 2004 http://wbec-ridderkerk.nl/html/UCIProtocol.html
77[01:03:28.670] lmabacus: 42
78[01:03:34.400] d0rus: 6 vs each
79[01:03:48.507] Aloril42: (script) if 1-0: All: 3861(+4), Kom: 3731(-5); if tcecDD : All: 3852(-5), Kom: 3741(+5); if 0-1: All: 3847(-10), Kom: 3746(+10); simulation and estimated start times: https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-90-speculation-simulation/
80[01:03:51.253] lmabacus: !reverse
81[01:03:51.636] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
82[01:03:51.961] sometimesithurts: how could one find how many positions are in cdb? the db info only contains db size in GB .. @atanh
83[01:03:54.192] tinduz_r: Thanks.
84[01:03:54.807] eval_bot Game #62 "KomodoMCTS 2381.00" vs. "AllieStein v0.5-dev_1359f44-n10": [book] 9. O-O Ne7 10. Qc2 h6 11. Re1 O-O-O 12. Rb1 Kb8 13. a4 Nb3 14. Nxb3 Bxa4 15. Nfd2 Nc6 16. Ra1 Bxb3 17. Nxb3 cxb3 18. Qd1 Na5 19. Be3 a6 20. Bf1 Nc4 21. Bxc4 dxc4 22. Qf3 Rd7 23. Ra4 Qb5 ... 0-1
85[01:04:20.309] atanh: It's a 3x DRR (double round-robin), so 6 games for each pair of engines.
86[01:04:50.948] alchemist888: Go KMC!
87[01:05:09.443] AngraMainyu2018: !gokomodo
88[01:05:13.856] lmabacus: NN Harry attack!
89[01:05:14.061] Nightbot: drezzZZZ
90[01:05:29.963] atanh: @sometimesithurts Good question. I don't know if there's a way to get that directly from CDB or not. Best bet is to ask @noobpwnftw directly if/when you see him in chat.
91[01:05:44.583] Aloril42: ScorpioNN has been a while above 90% for relegation.
92[01:05:45.892] AngraMainyu2018: !gomodo
93[01:05:46.117] Nightbot: sassKO
94[01:06:09.418] sometimesithurts: @atanh ok thx
95[01:06:43.287] atanh: Interesting divergent evals already. I don't think either KMC or Allie have any contempt setting, do they?
96[01:07:14.543] atanh: !cdb
97[01:07:15.827] chessdbcn (10w) • Qc2 4 ! (16-02) • Nf1 3 ! (24-01) • h4 3 ! (19-01) • g3 3 ! (21-01) • O-O 2 ! (18-02) • Kf1 0 ! (16-03) • pv Qc2 Ne7 O-O Qc7 Ra1 Qb6 a4 Nec6 • <ChessDBCN>
98[01:07:56.834] atanh: @sometimesithurts np :)
99[01:08:51.353] AngraMainyu2018: I don't know whether they have contempt. My guess is that they do not.
100[01:09:10.026] atanh: Oh no, this might be another win for Allie. More danger for Leela. I hope Leela can win directly against Allie at least once or twice. That would help a lot.
101[01:09:14.931] sometimesithurts: !time
102[01:09:16.163] eval_bot 79/168 games left • Average duration: 02:59:43 • Estimated division end: Tue, 24 Sep 2019 03:02:37 GMT • Shortest game: 00:59:55 • Longest game: 04:06:54
103[01:09:44.915] AngraMainyu2018: Kmc has a contempt setting, but I think it's set to 0
104[01:09:45.281] sometimesithurts: 168/8*2
105[01:10:12.630] atanh: Allie just optimistic as usual, then, I guess. :)
106[01:10:19.961] AngraMainyu2018: (I think, but I'm not sure)
107[01:10:29.894] sometimesithurts: = 42 games each
108[01:10:50.772] AngraMainyu2018: @atanh, I'm not sure it's optimism, or a different scale for the evals.
109[01:11:06.665] Y_h_w_h: mouseover the little King beneath Engine name -> current settings
110[01:11:20.776] haxboi25: allie is smart
111[01:11:24.160] Y_h_w_h: besides
112[01:11:30.215] AngraMainyu2018: AS has the old lc0 eval scale iirc, so it's not directly comparable with the others
113[01:11:42.305] Aloril42: @AngraMainyu2018 You can see Contempt is set to 0 by looking at KomodoMCTS options.
114[01:11:42.817] atanh: @sometimesithurts Yep, we're just past the half-way point in DivP, which was just after #84, when each engine had played 21 games.
115[01:12:04.450] AngraMainyu2018: @Aloril42, thanks. :)
116[01:12:32.120] mdt_sf: !db
117[01:12:33.297] eval_bot (11w) +0=0-0 • No moves • No games • <Lichess>
118[01:12:48.065] xoroshiro: !reverse
119[01:12:48.217] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
120[01:12:51.442] eval_bot Game #62 "KomodoMCTS 2381.00" vs. "AllieStein v0.5-dev_1359f44-n10": [book] 9. O-O Ne7 10. Qc2 h6 11. Re1 O-O-O 12. Rb1 Kb8 13. a4 Nb3 14. Nxb3 Bxa4 15. Nfd2 Nc6 16. Ra1 Bxb3 17. Nxb3 cxb3 18. Qd1 Na5 19. Be3 a6 20. Bf1 Nc4 21. Bxc4 dxc4 22. Qf3 Rd7 23. Ra4 Qb5 ... 0-1
121[01:14:15.441] atanh: @AngraMainyu2018 You're right, it's not actually optimism. I thought better only after posting that chat. It's actually simply due to the fact that its evals are derived from its estimated probability of winning, so a slight positive score just means that she thinks white has a slightly higher chance of winning. But it's still a close score, so chances of draw are still high.
122[01:16:49.313] atanh: !cdb
123[01:16:50.534] chessdbcn (11b) • Qc7 -2 ! (12-04) • Nc8 -7 ! (17-03) • a6 -10 * (16-06) • O-O-O -10 * (18-03) • pv Qc7 Nf1 O-O-O Bf4 Kb8 Bh2 Bb5 Ne3 • <ChessDBCN>
124[01:17:22.148] atanh: !c02
125[01:17:22.408] Nightbot: C02 French, Advance Variation
126[01:17:58.882] atanh: C02h French, Advance Harry Variation.
127[01:18:50.916] atanh: !db
128[01:18:52.288] eval_bot (12w) +0=0-0 • No moves • No games • <Lichess>
129[01:20:54.839] josephviruses: People still doubt alpha zero even if its results have been replicated by leela here.
130[01:22:56.760] josephviruses: !goallie
131[01:22:56.977] Nightbot: VoHiYo
132[01:26:07.633] lmabacus: Looks like 9. h4 scored the best out of book on 365chess but nearly everyone played 9. O-O
133[01:27:31.727] lmabacus: Like ElzChess said in his latest video, it's strange that h4 isn't played more in the French
134[01:27:47.076] Jeroen_TCEC: Wow, Allie in the winning mood in Div P!
135[01:29:37.895] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
136[01:29:57.175] alchemist888: This game reminds me this one https://www.tcec-chess. com/archive.html?season=14&div=sf&game=35
137[01:30:01.145] Jeydra: PowerUpL VoHiYo PowerUpR
138[01:32:49.820] darkcam17: SF would still be first rn if it didn't crash twice
139[01:37:40.484] atanh: Some interesting parallels in this game with that Leela vs SF game where she did the rook lift to h3, g3.
140[01:41:42.882] drdisentangle: drdisentangle subscribed with Twitch Prime. They've subscribed for 11 months! 
141[01:42:54.164] Jeydra: Looks like KMC has the same weaknesses SF does at playing the French
142[01:43:15.929] Qazcraft: what about the french causes SF to perform worse?
143[01:43:40.142] radjehuty: maybe nothing and leela is just very good against it?
144[01:44:54.173] f1tof5: Allie-LC0 final should be great.
145[01:45:22.313] Aloril42: !hypesub @drdisentangle
146[01:45:22.545] Nightbot: leeSUB thickL @drdisentangle thickR tcecBL tcecB tcecW sassHYPE scardSUB
147[01:46:47.815] lmabacus: KMC keeps putting off castling
148[01:47:59.738] josephviruses: RF is still at +0.5 even if the NN's are already >1
149[01:48:45.590] atanh: !cdb
150[01:48:46.847] chessdbcn (19b) • O-O-O -52 ? (24-01) • Qd7 -59 ? (19-01) • Rg8 -61 ? (22-01) • Ba4 -67 ? (15-01) • Qb6 -76 ? (01-01) • Nf5 -83 ? (01-01) • pv O-O-O Bh4 Kb8 Qb1 Re8 • <ChessDBCN>
151[01:48:59.616] tunnelr: allie about to sice another win?
152[01:49:26.608] Jeydra: @Qazcraft don't know, might be something to do with the pawn structure. I know Alayant mentioned that SF has some kind of king pawn shelter code that covers the f, g and h files, but in the French that shouldn't apply because castling Kingside gets you killd
153[01:50:47.428] lmabacus: At this point in the game I trust Allie and BL over RF.
154[01:51:28.581] LavaLamp18: would be impressive if AS wins both sides of this opening
155[01:53:13.784] VetustaMorla1970: Allie is not joking!
156[01:53:23.248] f1tof5: Imagine you're a top GM playing Allie or LC0. You get to this position as black, and they tell you you're pretty much lost. Say what?? Why?? Where did I go wrong? LUL
157[01:53:28.162] lmabacus: Probably an objective win for white at this point
158[01:55:52.257] twindlers: !polls
159[01:56:14.534] atanh: !cdb
160[01:56:15.734] chessdbcn (21w) • Bh4 103 ! (01-00) • pv Bh4 O-O-O Qb1 g5 fxg5 hxg5 Bf2 Nf5 • <ChessDBCN>
161[01:56:46.132] twindlers: !cbd
162[01:57:01.041] twindlers: !thc
163[01:57:08.666] twindlers: !lsd
164[01:57:12.017] Lord_Aldrich_: maybe 1...e6 is already dubious and 2....d5 loses outright?
165[01:58:41.259] lmabacus: Just checked the reverse game. So RF evaluated the opening as white +0.22. KMC played RF recommended move for the first ten moves, and at that point RF's eval was... -0.4
166[01:58:43.226] twindlers: or maybe KMCTS is closer to 100 elo below Allie than the rating system on TCEC would have you believe
167[01:59:39.864] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
168[01:59:55.984] Lord_Aldrich_: the pattern so far seems to be that AB engines, including SF, have no clue how to play the french as black
169[02:00:37.971] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
170[02:01:04.940] lmabacus: The French "octagon of doom" installed by Allie
171[02:01:18.612] atanh: I think it's more general that they have difficulty with closed positions overall, compared to NNs at least.
172[02:03:01.540] zjuwyz: SF has crashed?! Twice?! What happened?
173[02:05:03.309] gakuj505: 1-0 go go
174[02:05:28.608] zjuwyz: Why has SF crashed twice?
175[02:06:12.074] atanh: !cdb
176[02:06:13.324] chessdbcn (23b) • g6 -64 ? (05-01) • Kb8 -107 ? (12-01) • Ba4 -110 ? (17-01) • Nc6 -114 ? (02-01) • Kc7 -114 ? (16-01) • f6 -122 ? (02-01) • pv g6 Kh2 Nc6 f5 • <ChessDBCN>
177[02:09:46.833] lmabacus: Feels like 23. O-O might have thrown away the win
178[02:12:09.792] f1tof5: Based on Github discussion it now seems clear SF crashed because of faulty third-party DLL. But so it goes.
179[02:12:57.863] klnka: Still a win. White has a big advantage in the f, g, h columns. Rest is a fortress. g6 prevented an immediate powerful attack. But not enough.
180[02:14:21.526] STJukes: ah it's probably going to be Allie vs Lc0 in Sufi
181[02:16:53.444] gary_internet: Yeah even if SF doesn't crash for a 3rd time and automatically get relegated, it's going to be tough to beat Lc0 when the 2 crashes it's suffered so far will count against it in the tiebreaks. It's a shame really. Weirdly I'm now hoping that SF does crash a 3rd time so we can see it ruin League 1 next season.
182[02:17:01.591] atanh: !cdb
183[02:17:02.853] chessdbcn (27w) • Qd2 145 ! (18-00) • pv Qd2 Kb8 Rg1 Be7 Bxe7 • <ChessDBCN>
184[02:17:12.849] f1tof5: SF Github discussion is interesting. User CoffeeOne pointed out the problem on 10 July but it seems no one took enough notice.
185[02:19:21.386] STJukes: @gary_internet yeh I didn't think of that. A nice consolation prize seeing SF in div 1
186[02:19:24.492] lmabacus: Don't see how white makes any progress in this position. There's no open file for the rooks except the g-file, and black can just exchange the rooks there
187[02:21:21.569] f1tof5: Scorpio, Ethereal and Laser might "steal" a win off SF in next season's L1. But one would expect a record-breaking result by SF.
188[02:22:50.389] lmabacus: Of course, KMC is champion at snatching defeat from the jaws of equality
189[02:22:52.751] gary_internet: @f1tof5 If SF plays in League 1 next season I would expect it to finish in 1st place witohut losing a single game. The only question in my mind is, how few draws can it get.
190[02:23:13.678] f1tof5: Yep. Agreed
191[02:25:06.189] gary_internet: On Stefan Pohl's site, in a 1000 game sample, this version of SF played against Ethereal 11.53 and scored +679=311-10.
192[02:25:32.621] nickpelling: I don't think the Fish would be offering too many grandmaster draws.
193[02:25:43.810] gary_internet: So I don't expect Ethereal to be stealing any games. Nor Xiphos or Laser, or KMCTS, or anyone else.
194[02:26:24.900] f1tof5: No, but not completely unrealistic for that to happen.
195[02:27:18.135] gary_internet: True. Improbable, but not impossible.
196[02:27:56.716] makishiima_shogo: AS is the strongest engine in the world PogChamp
197[02:29:37.341] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
198[02:30:53.988] f1tof5: If all the major pieces come off, black puts bishop on e6 and pawns on a4 and b5 it would be a draw. So can Allie prevent that?
199[02:32:50.971] nickpelling: @makishiima_shogo luckiest, for sure. :-p
200[02:33:19.126] klnka: Bishop was necessary for a win. Draw now.
201[02:34:18.446] f1tof5: If Allie can control the g-file this is no draw, but not obvious to me how to do that.
202[02:34:52.228] klnka: She can not. Any decent player can prevent that.
203[02:35:04.114] sometimesithurts: !8ball is it a draw?
204[02:35:23.886] lmabacus: Winning move was 23. f5. BL's PV has white blasting through on the kingside
205[02:35:56.429] sometimesithurts: end of rf pv seems like a dead draw
206[02:36:05.349] atanh: !cdb
207[02:36:07.186] chessdbcn (31b) • Kc7 -116 ? (19-01) • a6 -123 ? (19-01) • Kb8 -123 ? (18-01) • Bc6 -124 ? (25-01) • Ref8 -131 ? (03-01) • Bb5 -133 ? (02-01) • pv Kc7 Rh1 Kc8 • <ChessDBCN>
208[02:37:12.443] klnka: Yes, Allie should attach before g6.
209[02:37:21.133] Jeroen_TCEC: Looks like a draw now
210[02:37:43.013] klnka: But I am not sure she did everything right after that.
211[02:37:51.176] Tesseract_A: this looks holdable
212[02:38:19.915] nickpelling: Don't underestimate kmcts's generosity in a slightly worse position. :-/
213[02:38:27.227] Tesseract_A: breaking points are definitely not concrete yet
214[02:38:30.664] f1tof5: Putting queen on h1 and trying to overload black's control of g8 and d5 may be possible? No? My worry though is Bc2->e4
215[02:39:08.036] Qazcraft: if every engine favors white and 3 of them by more than a pawn, it's not over yet
216[02:39:26.246] Tesseract_A: Bishop on e4 is not maneuverable
217[02:39:56.560] sometimesithurts: draw, unless allie's inflated eval, give her the incentive to push again for a lose Kappa
218[02:40:09.431] Tesseract_A: it's infront of the pawn structure but it might encourage a trade. not sure yet.
219[02:40:12.514] atanh: !cdb
220[02:40:13.748] chessdbcn (32w) • Rxg8 134 ! (05-00) • pv Rxg8 Rxg8 Qh2 Bc6 Rg1 • <ChessDBCN>
221[02:41:20.604] HenriKar_Chess: I am ... ever more aghast at the _size_ of that database!
222[02:41:41.852] atanh: Looks like any way you slice it, white is able to build up an ever-increading advantage. K side pawns structure broken, black bishop useless against white pawns.
223[02:42:18.355] nickpelling: It's just a Raspberry Pi with a randomised, don't be fooled. :-p
224[02:42:40.557] noobpwnftw: so what's the next goal? Go book? Kappa
225[02:42:48.941] nickpelling: *randomiser
226[02:43:29.035] f1tof5: Where is chessdbcn stored and how is it managed/updated?
227[02:43:53.458] noobpwnftw: stored on my server
228[02:44:01.628] lmabacus: Now it seems like Qa5 was a counter blunder. Not smart to pull the queen off the back rank and block a5
229[02:44:15.830] sometimesithurts: hi @noobpwnftw , how could one find the number of positions in db? the info only contains the db size in GB.
230[02:44:17.827] noobpwnftw: updates are automatic, no management needed
231[02:44:19.310] f1tof5: This position is looking won.
232[02:44:31.818] f1tof5: And what's it updated from?
233[02:44:41.902] noobpwnftw: self play
234[02:44:58.273] Stephane_Nicolet: @noobpwnftw Thanks for the machines :-)
235[02:45:05.416] Tesseract_A: i don't see the win....
236[02:45:15.386] A1ultra: Alliestein suddenly winning everything?
237[02:45:17.066] noobpwnftw: @sometimesithurts it is not possible to provide accurate number of positions due to LSM trees used
238[02:45:32.361] f1tof5: Which engines?
239[02:45:39.461] noobpwnftw: @Stephane_Nicolet it seems people can scale # of patches with cores well Kappa
240[02:45:42.386] HenriKar_Chess: How about an _in_accurate one?
241[02:46:00.814] lmabacus: Without the black queen on the back rank, suddenly white can control the g-file
242[02:46:06.070] noobs_killer: French , scam ,1-0!
243[02:46:22.185] sometimesithurts: what about an approximation ?
244[02:46:51.252] nickpelling: All the engines are now seeing lines with Qf6 and Rg6!
245[02:46:52.374] noobpwnftw: I will try to add it
246[02:47:09.577] A1ultra: Redfish is liking this
247[02:47:35.461] lmabacus: If the black queen was on d8 at this point, black can ignore the rook, but now it has to duck and the white queen worms her way in
248[02:47:44.788] sometimesithurts: wow thanks alot
249[02:48:17.062] fferen: dead won
250[02:49:11.022] noobpwnftw: "RocksDB can estimate number of keys through DB property “rocksdb.estimate-num-keys”. Note this estimation can be far off when there are merge operators, existing keys overwritten, or deleting non-existing keys." so it will be far off since it fits 2/3 of those criteria
251[02:50:03.854] Stephane_Nicolet: @noobpwnftw Indeed, only exception is when they are sleeping, and even it is not clear
252[02:50:45.281] Tesseract_A: wow Qd8 Rg6!
253[02:51:56.650] Tesseract_A: i didn't see that one
254[02:52:12.566] klnka: Still not a win.
255[02:52:24.701] nickpelling: It's why Qa5 was a mistake. :-(
256[02:52:30.267] Tesseract_A: it might not be a win but it's pretty neat
257[02:52:58.177] klnka: K played terrible for sure.
258[02:53:07.581] f1tof5: So Rg6 Qxf6 exf6 fxg6 hxg6 is actually winning for white????
259[02:53:13.443] noobpwnftw: @sometimesithurts 8 billion it seems now
260[02:53:52.017] sometimesithurts: wow it is a lot!!!
261[02:54:15.530] lmabacus: Rg6 Qxf6 Rxf6
262[02:54:33.940] tinduz_r: Black's LSB always getting stuck in a bad spot seems to be a recurring pattern in French against NNs.
263[02:54:42.036] nickpelling: And then the f5 pawn drops
264[02:55:01.440] gary_internet: Does anyone know of any up and coming NN engines? The only ones I am aware of that aren't in DivP right now are FireNN and Winter (which from what I understand is hybrid). Are there any others in the works?
265[02:55:37.676] Jeroen_TCEC: AB engines often have bad bishops. They don't understand that concept and put all their pawns on the same colour as their bishop.
266[02:55:37.873] chess_fan26: @gary_internet dark queen
267[02:55:40.391] Ove__: !chessfighter
268[02:55:40.641] Nightbot: Chess Fighter is an original neural network based engine written in C++ by Alexander Lim. http://talkchess.com/forum3/viewtopic.php?f=2&t=70206
269[02:55:46.222] lmabacus: K ready to blunder again
270[02:55:51.423] gilles_le_volatile: looks draw
271[02:56:33.826] chess_fan26: !Fatfritz @gary_internet
272[02:56:34.059] Nightbot: Fat Fritz runs on a modified version of the open-source Lc0 binary, but using its custom-trained Fat Fritz neural network. https://en.chessbase.com/post/fat-fritz-what-on-earth-is-that https://en.chessbase.com/post/using-fat-fritz-in-the-engine-cloud
273[02:56:48.007] Jeroen_TCEC: Replay some SF games in Div P. It has a bad bishop quite often. Puts its own bishop behind a wall of pawns. NN engines understand this concept a lot better
274[02:56:50.635] klnka: K may blunder and allow a bishop sac win. Otherwise dead draw.
275[02:56:59.316] gilles_le_volatile: At least I don't see how white progress
276[02:57:02.552] f1tof5: @lmabacus Ah yes, thanks.
277[02:57:11.741] Qazcraft: well every single engine does....
278[02:57:27.301] gary_internet: @chess_fan26 Good shout. @Ove__ Thanks, yes forgot about that one! Just trying to workout if anything will prevent the 2 relegated engines (whichever they end up being) from promoting straight back to DivP next season.
279[02:57:38.863] Ove__: chessfighter is in QL
280[02:58:15.117] HenriKar_Chess: Would Fat Fritz satisfy uniqueness requirement?
281[02:58:39.146] gary_internet: I'm working on the basis that only an NN could make rapid enough progress in development to climb through all the divisions and maybe stop the 2 from DivP going straight back up.
282[02:59:22.958] f1tof5: Rg7 followed by Bc2 looks compelling.
283[02:59:39.234] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
284[02:59:40.170] gilles_le_volatile: Bad / Good bishop is a tricky concept because very often you can activate your bad bishop , I remember a good example of htis in silman book "how to reassess your chess"
285[02:59:41.397] chess_fan26: Even ScorpioNN is a NN
286[02:59:58.391] gary_internet: The likes of Ethereal, Laser, Xiphos, Andscacs, Fizbo and Jonny have basically ceased development. I suppose we might see ChiroNN?
287[03:00:36.790] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
288[03:00:41.948] lmabacus: Wonder what's the difference between a6 and a5 here.
289[03:01:06.553] gary_internet: @chess_fan26 Yes, but I'm counting ScorpioNN as one of the 2 engines from DivP that is going to be relegated. I think by next season ScorpioNN will be a little stronger than it is now, and it's already made it to DivP.
290[03:01:53.020] chess_fan26: Even I think it will make it back next seadon
291[03:03:19.813] atanh: @noobpwnftw What are your own personal goals or hopes for what you'd like to see come about? E.g. Are you interested in getting some more people using CDB? Would you like to turn CDB into a guest entry for next season's QL (maybe as a bonus event)? Something completely different? I could think of lots of things (brainstorming), but it's best if it's something you're personally interested in. :)
292[03:08:11.954] nsousa254: COME ON SF BEAT THOSE SCAMMERS
293[03:09:12.237] tunnelr: at least rf doesnt show draw
294[03:11:58.581] Checello: Is there an undestandable explanation of allies strength?
295[03:13:17.666] tinduz_r: !allie
296[03:13:17.841] Nightbot: Allie is a neural network based engine written by Adam Treat, inspired by the AlphaZero paper and the Lc0 project. See - https://github.com/manyoso/allie , https://www.patreon.com/gonzochess75 and http://www.chessdom.com/alliestein-the-new-neural-network-entering-tcec-s15/ , Join the discord channel to follow Allie - https://discord.gg/UnbsJRU
297[03:13:23.006] Dabbler1: can allie be stopped??
298[03:13:24.679] OverUsedChewToy: Oh boy, Allie winning! this is scandalous MrDestructoid / PopCorn
299[03:13:30.450] lmabacus: Black's position seems impenetrable enough that even K can't mess it up, though I'm sure it'll try its best
300[03:13:59.410] klnka: RF eval is wrong. K needs to make a mistake. Only thing to do is Bishop sac.
301[03:14:50.254] Tesseract_A: @OverUsedChewToy scamalous*
302[03:15:33.638] Dabbler1: kmcts not doing itself any favors with the clock situation
303[03:16:04.625] lmabacus: Yeah, it and H are the two slowest engines here
304[03:16:38.438] Dabbler1: with an important difference being that houdini is extremely solid :D
305[03:16:42.341] Tesseract_A: RF seeing something new?
306[03:17:35.968] Checello: zhx nightbot, I read this, but I think it is very very hard to undestand.
307[03:18:07.457] klnka: Nothing. They just don't see the semi fortress.
308[03:18:56.513] klnka: If black moves the b pawn then something changes. But even K will not do that.
309[03:18:57.570] tinduz_r: An intuition on why Allie is strong = think Leela evals but with Minimax AB search.
310[03:20:11.399] noobs_killer: Allie= Leela+Stockfish, endofleelaera!
311[03:20:22.672] Dabbler1: 3 wins with black when the next most is only 1. Very impressive
312[03:20:56.611] tinduz_r: It also abandoned "zero" I think, hoping to get some improvement.
313[03:20:57.632] HenriKar_Chess: !horizone
314[03:20:57.866] Nightbot: Looks like we hit the horizontal zone --L0RINDA
315[03:21:33.977] Dabbler1: lol
316[03:21:48.887] OverUsedChewToy: i think an allie - lc0 final would be compelling
317[03:21:54.746] fresh_salsa_chips: @tinduz_r also because stockfish gave it a crash point Kappa
318[03:22:24.880] drgibbs_: leela team was never trying to make the best chess engine they could, they were trying to make the best they could do under a set of restrictions
319[03:22:55.294] drgibbs_: it isn't surprising that someone can do better without the self-imposed restrictions
320[03:22:56.060] Dabbler1: i am warming up to the idea of that final. i wanted a rematch but allie deserves one of the spots if the tournament ended right now
321[03:23:03.204] fresh_salsa_chips: "leela team was never trying to make the best chess engine they could" questionable claim CoolStoryBob
322[03:25:06.557] Occyroexanthub: no he is right
323[03:25:08.798] drgibbs_: leela team are trying to re-implement the alpha zero paper with the same zero human influence idea
324[03:25:15.397] tunnelr: so allie is leela and sf's offspring?
325[03:25:16.257] noobs_killer: Lela One>> Leela Zero.
326[03:25:18.738] Occyroexanthub: they tried to do the best WITH rhe restriction to be always zero
327[03:25:30.695] f1tof5: Where's the result of the vote by divp devs if SF could change its DLL?
328[03:25:42.395] zeekza: the evals are really flat
329[03:25:50.237] drgibbs_: it is an experiment, not an attempt at making the strongest engine -- trying to reproduce results of the closed source google claims
330[03:26:46.741] zeekza: why is the slowmode raised to 30s from the old 10s? it's a pretty long time and makes it a bit akward to have a conversation
331[03:27:25.821] tinduz_r: You read for 10s, then type response for 20s. :P
332[03:27:32.072] sirinial64: @zeekza subscribe ! Kappa
333[03:27:44.385] zeekza: cashmoney machine broke
334[03:27:45.086] Occyroexanthub: yeah ti sucks big time. that's why you should subscribe :P
335[03:28:02.337] Ove__: !vote
336[03:28:02.566] Nightbot: ""Should Stockfish team be allowed to replace libwinpthread-1.dll with non-buggy version?" AntonMihailov: The final results of the voting are now in. Out of 8 votes we have 6 "yes", 1 "abstain", and one "no". This final no is the veto. Voting is anonymous and TCEC team will respect that and the outcome of the authors' vote nadedKing
337[03:28:06.756] bromvlieg12: !endofanera
338[03:28:06.971] Nightbot: End of era has finally arrived. tcecNE has started™ see https://en.wikipedia.org/wiki/End_of_an_Era
339[03:28:38.177] fresh_salsa_chips: 6 yes, 1 Houdart, and one no = no KappaPride
340[03:28:39.670] zeekza: it's hard to subscribe when ur a broke college stuent
341[03:28:53.926] tunnelr: subscribe with prime
342[03:28:56.610] bromvlieg12: Guys, will Leela miss out in Superfinal?
343[03:29:04.280] Occyroexanthub: maybe
344[03:29:10.919] zeekza: I don't have prime bc m a broke college student as stated
345[03:29:15.224] Occyroexanthub: right now seems that way
346[03:29:21.341] Occyroexanthub: but 19-20 games remain
347[03:29:31.129] sometimesithurts: moat likely , unless sf DQ Kappa
348[03:29:36.376] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
349[03:29:37.166] tunnelr: well nows a good time to subscribe it's less than $3
350[03:29:49.309] f1tof5: @fresh_salsa_chips Thanks. Houdart obviously doesn't care if people don't like him. LUL But I think his vote is the fair one.
351[03:30:30.601] fresh_salsa_chips: @f1tof5 abstinence, the best policy Kappa
352[03:30:32.903] Jeydra: How we even know Houdart is the one who voted "no"?
353[03:31:00.822] fresh_salsa_chips: @Jeydra isn't houdart the "abstain" guy?
354[03:31:04.548] sirinial64: Houdart is most likely abstention
355[03:31:36.750] sirinial64: No vote is either komodo or stoof author
356[03:31:37.679] zeekza: wait so there's a bug in the current ver of sf?
357[03:32:05.124] HenriKar_Chess: !dll zeekza
358[03:32:05.387] Nightbot: ""Should Stockfish team be allowed to replace libwinpthread-1.dll with non-buggy version?" AntonMihailov: The final results of the voting are now in. Out of 8 votes we have 6 "yes", 1 "abstain", and one "no". This final no is the veto. Voting is anonymous and TCEC team will respect that and the outcome of the authors' vote nadedKing
359[03:32:35.180] zeekza: what's the bug tho/ when was it found and how likely is it to occur
360[03:32:44.898] HenriKar_Chess: !74
361[03:32:45.134] Nightbot: Crash unless crash dump investigation says reason is not in tcecSF submission. See !51 and !dll
362[03:32:46.568] sirinial64: !51
363[03:32:46.723] Nightbot: In absence of further evidence, and as the logs didn't provide counter evidence, TCEC has no choice but to rule this as a tcecSF crash. See also https://github.com/official-stockfish/Stockfish/issues/2291 , https://github.com/official-stockfish/Stockfish/issues/2229 , and the Crash info tab - see also !74 and !dll
364[03:32:58.919] f1tof5: !dll
365[03:32:59.174] Nightbot: ""Should Stockfish team be allowed to replace libwinpthread-1.dll with non-buggy version?" AntonMihailov: The final results of the voting are now in. Out of 8 votes we have 6 "yes", 1 "abstain", and one "no". This final no is the veto. Voting is anonymous and TCEC team will respect that and the outcome of the authors' vote nadedKing
366[03:33:08.000] f1tof5: !74
367[03:33:36.289] zeekza: wait so does that mean SF is DQed bc of the crash?
368[03:33:44.126] AfuroZamurai: @zeeksa sometimes SF just won't return a move (so far 2/20 games)
369[03:33:47.868] f1tof5: Only if it crashes one more time.
370[03:33:51.986] lmabacus: Only if it crashes a third time
371[03:33:55.681] f1tof5: But that seems likely.
372[03:33:56.021] atanh: @gary_internet Well, Laser has made amazing progress recently. It's roughly on par with Ethereal now. Perhaps those two, devs Laldon and AGE who have worked together, will start competing between themselves to push their engines into divP. They're both right on the cusp.
373[03:34:03.780] HenriKar_Chess: @zeekza not yet, but at the next one it'll be
374[03:34:12.855] zeekza: but also #crashes is the first tie breaker too right
375[03:34:20.670] lmabacus: Right
376[03:34:43.592] sirinial64: @atanh gonna be tough if sf is in league 1 next season LUL
377[03:35:00.315] zeekza: wait so does it like just analyse until time runs out (lock) or halt entirely
378[03:35:02.306] f1tof5: I wonder if reducing SF threads would reduce risk of the bug? Presumably it goes into deadlock.
379[03:35:07.241] drgibbs_: is a crash a loss or a rematch?
380[03:35:20.090] f1tof5: But the ignominy of SF having to reduce its threads. :(
381[03:35:21.145] fresh_salsa_chips: crash is a loss
382[03:35:21.481] atanh: What would happen with a b3 or b4 push from white?
383[03:35:29.308] lmabacus: Though given SF's performance sans crashes, it should finish ahead of the other two (or least Leela) if a third crash isn't forthcoming.
384[03:35:35.822] HenriKar_Chess: @drgibbs_ if it's a real crash (not just a network disconnect) it's a loss
385[03:36:05.343] zeekza: is it just sf entering a deadlock or does it actually halt too?
386[03:36:28.344] HenriKar_Chess: @atanh HF thinks b3 is 0.00 and b4 is NE-GA-TI-VE
387[03:36:30.202] tinduz_r: It responded to a "stop" command, they said.
388[03:36:58.762] zeekza: what's HF?
389[03:37:09.970] f1tof5: Can one argue that SF hasn't crashed, but lost on time? Then disq. isn't on the cards?
390[03:37:10.030] lmabacus: Homefish
391[03:37:20.588] Jeydra: Hmm, draw?
392[03:37:23.623] lmabacus: Stockfish on someone's home computer
393[03:37:30.652] zeekza: yea ye
394[03:37:41.253] Jeroen_TCEC: Draw, unless black blunders
395[03:37:53.934] Jeydra: I don't see what White can do if Black just shuffles with the king
396[03:38:08.704] zeekza: why is RF's evals so constant
397[03:38:08.910] tinduz_r: KMCTS never blunders. Kappa
398[03:38:10.377] lmabacus: It didn't respond to commands after time ran out, so it's considered to have crashed
399[03:38:13.755] ma_d____m_an: It look fortressy
400[03:38:33.876] Jeydra: I suppose White could sac an exchange on f7
401[03:38:52.589] lmabacus: RF has a flat non-zero eval when it encounters fortress type positions.
402[03:38:56.339] Jeroen_TCEC: All white's pieces are superior to black's, yet there is no way to crack black's defensive position. Interesting to see that even the top engines still do not see it.
403[03:39:03.575] zeekza: oh yea yep
404[03:39:06.715] HenriKar_Chess: @atanh but it changes a lot... for example after the current move it's b4 0.00 and b3 +1.00
405[03:39:09.547] alchemist888: We have 2 approaches and still can't detect many fortresses. Chess is too hard
406[03:39:27.289] HenriKar_Chess: The PV is 1.60
407[03:39:27.991] f1tof5: @Jeroen_TCEC You don't reckon f7 exch sac is possible?
408[03:39:33.970] zeekza: what are the 2 approaches?
409[03:40:08.617] tinduz_r: NN self play eval vs expert system?
410[03:40:08.784] Nightbot: drezzZZZ
411[03:40:19.313] alchemist888: NNs and AB HC engines of couse @zeekza
412[03:40:43.170] zeekza: oh i thought you meant like algorithms specifically for detecting fortesses
413[03:41:13.903] zeekza: shouldn't the eval zero by the 50 move rule?
414[03:41:22.899] alchemist888: I know a special algorithm for detecting fortresses - human
415[03:41:44.152] lmabacus: Oh, one of the engines will find a capture or pawn push by that point usually
416[03:41:46.122] tinduz_r: One may also think of the positive eval as more chances for black to mess it up.
417[03:41:52.836] HenriKar_Chess: @zeekza only if those 50 moves there are no takes and pawn moves
418[03:42:14.286] HenriKar_Chess: @zeekza and neither white nor black thinks with depth 50 anyway
419[03:42:16.387] zeekza: I mean to identify when a non zero engine eval is zero that takes a human... but sometimes humans might say its nothing but the engine sees a way to crack it
420[03:42:16.666] lmabacus: Most likely Allie will initiate a rook or queen exchange
421[03:42:17.655] f1tof5: Except many human players would try the f7 exch sac. I'm not sure whether it's sound but in human play it has a good chance of succeeding.
422[03:42:45.184] HenriKar_Chess: HF likes Qe7... any ideas?
423[03:42:58.052] zeekza: black can always stop the f7 sac tho
424[03:43:03.797] HenriKar_Chess: Well, ehm, previous move, not this one
425[03:43:06.933] lmabacus: Qe7 followed by what?
426[03:43:16.332] tunnelr: Qc5+?
427[03:43:38.312] gary_internet: Redfish hasn't gone above +1.87 for a long time.
428[03:43:49.038] Occyroexanthub: or below
429[03:43:52.122] atanh: @sirinial64 I'm still hopeful that SF won't crash a third time. But you're right, if it does, there'll be only one slot remaining to promote from L1.
430[03:43:55.517] tinduz_r: !horizone
431[03:43:55.734] Nightbot: Looks like we hit the horizontal zone --L0RINDA
432[03:43:56.488] zeekza: I mean I feel like the only benefit of Qe7 would be stopping Qe8 and Qf8 which would prevent the exchange sac
433[03:44:30.612] zeekza: wait what's this
434[03:44:30.957] lmabacus: What just happened?
435[03:44:31.933] atanh: Here we gooooooo!
436[03:44:44.347] tinduz_r: Q exchange?
437[03:44:46.876] HenriKar_Chess: Meh .. I just tried 54. Qe7. Ka7 55. Qc5+ Qxc5 dxc5 with crashing eval (1.6-> 0.9)
438[03:44:52.776] Occyroexanthub: it was inevitable K MCTS to blunder
439[03:45:00.175] Occyroexanthub: 9 losses so far
440[03:45:01.004] zeekza: !boom
441[03:45:01.189] Moobot: the concept of the "boom" is a TCEC invention [a BOOM means "eval explodes away from 0.00" - the opposite is a MOOB which therefore means "eval implodes back in the general direction of 0.00"]
442[03:45:05.981] Occyroexanthub: too bad this divP
443[03:45:11.015] alchemist888: bluuuunder
444[03:45:14.465] josephviruses: KMCTS gave up the only reason it is still defending this game
445[03:45:31.243] HenriKar_Chess: !boom3
446[03:45:31.478] Nightbot: twitchRaid shaggyNUKE shaggyNUKE shaggyNUKE
447[03:45:46.929] zeekza: why did komodo do that tho?
448[03:45:58.271] Occyroexanthub: now if SF wins then Leela 2 full points behind Allie and 1 point behind SF
449[03:46:01.218] lmabacus: It hates winning
450[03:46:19.047] Occyroexanthub: sufi more or less decided. even though ok 19 games remaining so a little premature to say that :P
451[03:46:20.247] josephviruses: I don't think Allie in the SuFi is going to be boring at all. I think Allie is going to be aggressive AF
452[03:46:21.460] zeekza: lmao, so it's an anti-giri then
453[03:46:35.235] HenriKar_Chess: @zeekza you never know, with an engine, not really, and even less when it's using some Monte Carlo method
454[03:46:38.966] m7md_wfa: aille vs SF sufi
455[03:46:42.144] Jeydra: Gotta play b3
456[03:47:23.400] tinduz_r: How to break?
457[03:47:30.830] klnka: Now it is possible to play with b pawns. Fortress is no more.
458[03:47:37.330] zeekza: this still isn't easy to break
459[03:47:45.984] HenriKar_Chess: @Jeydra HF likes b3 (in 3 plies)
460[03:47:47.769] sometimesithurts: so sf championship again
461[03:48:06.905] josephviruses: B3 or b4
462[03:48:10.605] mannimond: still enough time for a 3rd sf crash
463[03:48:12.713] tunnelr: kmcts is going to blunder again with time pressure
464[03:48:14.098] klnka: It is over.
465[03:48:33.367] zeekza: oh was it bc time pressure
466[03:48:40.153] atanh: @tinduz_r Actually, no, it *didn't* repsond to a 'stop', which is why it's considered a 'crash' not just a time loss. Actually it was a hang (deadlock), not a process termination (crash), but it's the same result here ('crash'). :)
467[03:48:45.009] klnka: b, c, d pawns attack.
468[03:48:59.621] gary_internet: Lc0 is ahead of SF on both crashes and direct encounter. I don't think SF will make the superfinal.
469[03:49:20.769] atanh: @HenriKar_Chess Thanks for that answer about pushing b pawn. :)
470[03:49:29.283] ladaova: Who voted "NO" ???
471[03:49:39.931] zeekza: it's anonymous
472[03:49:41.749] tinduz_r: @atanh Oh. Must have read a comment before wrongly. Thanks.
473[03:49:49.091] josephviruses: @gary_internet but SF needs to do is not crash and keep on winning. Leela hates to win.
474[03:50:32.819] zeekza: also like rules are rules so I kinda agree with that. Like I mean it would be nice to see SF at full strength for the spirit of things, but the rules say that you can't make changes once it starts so
475[03:50:40.170] josephviruses: Only thing that's keeping leela in top 2 is that it's not losing
476[03:51:00.649] atanh: Suppose B goes to f3, then push b3. If cxb3, K mops up the pawn. If black doesn't take the bait, then bxc4, and presumably dxc4, opening up diagonal for B on f3. Would it work?
477[03:51:03.693] zeekza: I mean not losing and winning are equally important
478[03:51:13.980] tinduz_r: Leela playing "too perfect" chess. Never loses, but may not win as much.
479[03:51:21.872] chess_fan26: I believe firmly if sf does not crash again then it can easily make it to the Sufi.
480[03:51:35.603] zeekza: how does the L bishop go to f3 (a dark square) @atanh
481[03:51:38.382] josephviruses: Leela is playing 'boring' chess atm
482[03:51:39.887] sometimesithurts: too perfect LUL
483[03:51:56.372] K_recreant: # of wins is most telling in assessing engine strength
484[03:51:58.576] josephviruses: Leela needs to learn her zero version of contempt
485[03:52:12.378] Occyroexanthub: not really
486[03:52:15.506] sometimesithurts: that's why she is throwing winning end games Kappa
487[03:52:24.322] zeekza: wait nvm I thought you said f6 I'm dumb
488[03:52:29.216] Occyroexanthub: number of points give the engine strength
489[03:52:36.795] josephviruses: H2h result is most telling of engine strength
490[03:52:45.062] Occyroexanthub: no purpose if you win but you lose a lot too
491[03:52:50.257] K_recreant: I disagree
492[03:52:50.725] HenriKar_Chess: @josephviruses looks more like Leela unlearnt some contempt and is playing more safely, but also less interestingly
493[03:53:00.037] skembris: bishop going to a4 now?
494[03:53:09.873] Occyroexanthub: well in h2h matches LEela is the best. but here is divP :P
495[03:53:29.901] chess_fan26: @Occyroexanthub its sad that leela can't make it to Sufi even after sending the strongest version
496[03:53:32.174] josephviruses: @HenriKar_Chess that's my observation too
497[03:53:33.288] viktprog: why allie and lc0 have similar % but their scores are very different?
498[03:53:35.141] zeekza: I mean zero losses means that Leela isn't losing any H2H
499[03:53:37.586] Jeydra: g
500[03:53:55.881] Occyroexanthub: @viktprog because of TCEC's fault
501[03:54:02.665] Occyroexanthub: of not updating formula for Leela's evals
502[03:54:04.237] ValentinaMelly: H
503[03:54:19.722] skembris: black running out of moves?
504[03:54:33.507] sometimesithurts: is it also Cato's fault? SeemsGood KappaRoss
505[03:54:45.960] HenriKar_Chess: !catofault
506[03:55:02.255] zeekza: well some engines tend to give different strengths of evals
507[03:55:08.925] josephviruses: How to choose the best net that tops the competition and not just SF?
508[03:55:37.264] Occyroexanthub: they should include contempt to LEela. but they don't as it's non zero
509[03:56:11.422] josephviruses: Contempt can't be learned, can it?
510[03:56:14.290] m7md_wfa: black running out of time
511[03:56:19.445] HenriKar_Chess: can you actually learn implicit contempt in a self-play scenario?
512[03:56:22.019] josephviruses: Or else leela should have it by now
513[03:56:32.430] sometimesithurts: I wish quick K-stoof draw..looking forward to see the sf-leela game
514[03:56:39.929] zeekza: I mean there is the 50 move rule which kinda acts like contempt
515[03:56:48.665] josephviruses: 42482 seems to have the best contempt that I saw in leela nets
516[03:57:35.861] atanh: KMC is toast with this time pressure.
517[03:57:37.632] gary_internet: KMCTS is getting ever closer to entering bullet mode.
518[03:57:41.745] HenriKar_Chess: @zeekza I've been wondering what you just phrased right now
519[03:58:18.206] josephviruses: The latest nets like to simplify very fast and try to rely on the other engine blundering in endgames, imo
520[03:58:19.196] HenriKar_Chess: in combination with 3-fold
521[03:58:22.479] zeekza: basically by the 50 move rule, doing nothing is zero and making a pawn move that continues being slightly positive and just draws slower would be preffered
522[03:58:22.961] tinduz_r: !0 from KMCTS
523[03:58:23.109] Nightbot: ZE-RO
524[03:58:51.951] HenriKar_Chess: !2fold maybe?
525[03:58:57.112] zeekza: and it changed it's mind
526[03:58:59.932] noobs_killer: !0.00
527[03:59:00.132] Nightbot: StoicFish tcecSF : Move along, nothing to see here.
528[03:59:13.982] josephviruses: That is why Leela's wins came from KMC mostlu
529[03:59:21.846] josephviruses: Mostly
530[03:59:35.289] lmabacus: Not entirely sure what white's plan should be, and it seems neither does RF
531[03:59:38.934] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
532[03:59:49.455] josephviruses: Leela likes winning against the toughest opponents of course, like SF
533[03:59:52.920] m7md_wfa: it is only 2 wins ?!
534[03:59:55.576] lmabacus: It keeps pulling the black rook off of the defense of the h7 pawn for no reason
535[04:00:06.994] klnka: None of these machines understand the situation.
536[04:00:11.512] lmabacus: h6 pawn*
537[04:00:27.090] Occyroexanthub: they already found the win for white what are you talking about? :P
538[04:00:36.109] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
539[04:00:42.180] Occyroexanthub: follwo some of the PVs
540[04:00:46.873] Occyroexanthub: follow
541[04:00:47.220] josephviruses: Allie will wait for the 50-move rule before kicking off b3
542[04:00:56.812] zeekza: oof
543[04:01:08.643] Occyroexanthub: black can't protect both f5 and f5
544[04:01:11.535] Occyroexanthub: huh
545[04:01:12.088] jigglefart: WHAT is even happening?!!?!?
546[04:01:16.399] Occyroexanthub: f7 and f5 i meant NotLikeThis
547[04:01:45.914] jigglefart: i dont see how either f5 or f7 are accessible though
548[04:01:48.011] Bossieman: b2-b3 and the infiltrate with king
549[04:01:55.408] zeekza: ^
550[04:02:07.447] klnka: Allie is holding back the winning move because she only cares immediate eval.
551[04:02:17.760] tunnelr: rook checks though
552[04:02:20.945] josephviruses: B3 or white sacrifices the Bishop for f5 and f7 pawns
553[04:02:26.504] zeekza: Ba4444 that's progresssss
554[04:02:34.566] lmabacus: I don't think I've seen a game with two separate sky-high RF horizones
555[04:02:57.847] zeekza: that's allie for ya ig
556[04:03:05.455] josephviruses: Those are rice terraces that can be found in Southeast asia
557[04:03:16.833] kuronitySE: Rg2 NotLikeThis
558[04:03:27.640] tunnelr: !bulletzone
559[04:03:30.001] zeekza: how will allie handle the time pressure?
560[04:03:32.537] jigglefart: i dom't know if b2-b3 is real infiltration though. cuz black rook can constantly attack
561[04:03:46.536] josephviruses: If SF v Allie in SuFi, we will have the most number of converted games
562[04:04:06.757] tunnelr: king can march up the black diagonal
563[04:04:24.470] tinduz_r: Allie is channeling he trolling powers of early Leela.
564[04:04:39.252] jigglefart: @tunnelr but without being checked by black rook???
565[04:04:39.752] zeekza: lol what did early leela do
566[04:05:09.742] jigglefart: not so sure
567[04:05:20.494] atanh: !boom4
568[04:05:20.696] Nightbot: nadedBomb pvpBoom pvpBoom pvpBoom pvpBoom
569[04:05:27.361] Bossieman: !boom3
570[04:05:27.501] Nightbot: twitchRaid shaggyNUKE shaggyNUKE shaggyNUKE
571[04:05:31.263] zeekza: UHHHHHHHHH maybe that won't be relevant lmao
572[04:05:34.607] HenriKar_Chess: @zeekza take a _long_ time to actually convert a good position
573[04:05:56.847] lmabacus: d8 square has been a source of misery for K three steight times now
574[04:06:03.323] HenriKar_Chess: !boom8
575[04:06:03.459] Nightbot: quackBomb jepsonHorn jepsonHype
576[04:06:07.807] Bossieman: Allowing Rh7 --> game over
577[04:06:08.854] tunnelr: ok i guess thats the gg
578[04:06:09.597] zeekza: what was the first time
579[04:06:23.221] lmabacus: Moved Q off of d8, moved queen TO d8, and moved rook to d8
580[04:06:45.994] tinduz_r: There was a funny video commentary of Leela's trolling against Fire in CCCC.
581[04:07:00.074] sometimesithurts: KMC on 0.00
582[04:08:01.811] tinduz_r: Early leela hijinks were very funny and sometimes frustrating, but funny.
583[04:08:12.820] tunnelr: damn allie knows its stuff
584[04:09:01.204] atanh: @josephviruses Zero version of contempt would be easy if they trained Leela against not just her latest self, but some previous versions of herself, presumably weaker. Then, she could, in theory anyway, learn to spot a weaker opponent via the 8 move history. Then she could, in theory, adjust the probabilities of moves based on slight differences in play. Voila! Contempt!
585[04:09:33.236] HenriKar_Chess: @atanh but is this still 'zero' ?
586[04:09:46.136] AfuroZamurai: wow, KMCTS self-destructing right and left...
587[04:09:49.664] atanh: !winrule 's coming!
588[04:09:49.939] Nightbot: Game ends in a win if there is a mate, engine resigns, time runs out, there is a crash, by 6-men TB adjudication, or by the TCEC winrule: both playing engines have an eval of at least 10.00 pawns (or -10.00 in case of a black win) for 5 consecutive moves, or 10 plies - this 10/10 rule is in effect as soon as the game starts. In the website this rule is shown as "TCEC win rule" with a number indicating how many plies there are left until it kicks in.
589[04:09:57.659] HenriKar_Chess: if you pit an engine against a weaker version of itself on purpose?
590[04:10:32.953] Bossieman: That is done all the time in training
591[04:10:57.202] HenriKar_Chess: @Bossieman I was not questioning the method, I was just wondering
592[04:11:00.701] m7md_wfa: Leela or SF should beat Aille If both want to be in sufi
593[04:11:30.996] drgibbs_: komodo doesn't want to resign
594[04:11:38.380] zeekza: lol
595[04:11:52.978] that_makes_sensei: KMCTS playing for Ka3
596[04:11:59.234] Bossieman: GG
597[04:12:51.317] zeekza: tfw komodo showed 0 3 pawns down
598[04:12:52.462] gakuj505: good game :3
599[04:13:05.555] drgibbs_: haha, convinced it was a drawing position
600[04:13:11.715] abwolffang21: KMCTS getting optimistic and thinking it was 0.06...
601[04:13:23.092] lmabacus: To be fair, K was right when that happened against Leela
602[04:13:39.555] zeekza: lmao, more like 0.06x10e5
603[04:13:40.547] chmariach: was this loss forced, or a KMCTS blunder? (time trouble?)
604[04:13:48.087] 1coghoti: @m7md_wfa what you said
605[04:14:02.809] tunnelr: probably a blunder
606[04:14:09.204] lmabacus: Blunder, don't think time trouble was even needed for them
607[04:14:14.438] zeekza: blunder almost surely, KMCTS had a fortress
608[04:14:18.615] chmariach: yikes
609[04:14:34.065] Bossieman: 12 ... Nb3 was the start of the fall
610[04:14:44.646] HenriKar_Chess: 72.. Rd8? was a decisive bad move
611[04:14:46.447] lmabacus: Qa5 first bad move, Qd8 second bad move, Rd8 final dagger
612[04:14:47.278] jigglefart: this looks drawn.
613[04:14:55.744] atanh: !hypealliestein
614[04:14:55.928] Nightbot: fl0mHype bueRocket VoHiYo fl0mHype
615[04:14:57.049] gakuj505: NotLikeThis
616[04:14:58.067] zeekza: lol
617[04:15:02.581] fresh_salsa_chips: rook and pawn ending, dead draw KappaPride
618[04:15:17.979] Aloril42: (script) 90: 1. AllieStein 3861(+4) - 7. KomodoMCTS 3731(-5) tcecW - tcecRIP https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-90-alliestein-v0-5-dev_1359f44-n10-komodomcts-2381-00/
619[04:15:25.271] tunnelr: gg winrule
620[04:15:28.238] gakuj505: ResidentSleeper KappaRoss Keepo
621[04:15:39.039] zeekza: Qd8 was also decisively a blunder as well
622[04:15:49.062] that_makes_sensei: I would queen every one of em and give stalemate in this position
623[04:16:01.579] tinduz_r: Allie 2 points comfortably ahead of 2nd.
624[04:16:08.012] Aloril42: (script) if 1-0: Kom: 3818(+7), Sto: 3758(-1); if tcecDD : Kom: 3808(-3), Sto: 3760(+1); if 0-1: Kom: 3801(-10), Sto: 3761(+2); simulation and estimated start times: https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-91-speculation-simulation/
625[04:16:16.715] zeekza: lol
626[04:16:27.449] m7md_wfa: Aille 3 wins ahead of everyone
627[04:16:36.595] atanh: !hypekomodo Komodo gets its revenge for its brethren!
628[04:16:36.820] Nightbot: KomodoHype actiHype knjRage knjLove
629[04:16:57.721] chmariach: allie even on par with bugless SF
630[04:16:58.669] lmabacus: SF would still win tiebreak against Allie if it hadn't crashed, but amazing that Allie's even tied at all.
631[04:17:15.657] zeekza: allie has 3 wins with black tho like that's more than all other engines in divp combines
632[04:17:33.762] frakty_zero: is it SSS or is allie really thay much bettet at defeating weaker oponnents than leela and if so why?
633[04:17:40.538] lmabacus: 81.9% chance of Allie making Sufi!
634[04:17:48.560] zeekza: whats sss
635[04:17:55.174] lmabacus: !sss
636[04:17:55.405] Nightbot: gopCAFE small sample size lol
637[04:17:56.035] Bossieman: !sss
638[04:18:19.980] zeekza: ah yes
639[04:18:38.352] HenriKar_Chess: Looks like Allie still has some of that implicit contempt :)
640[04:19:07.870] atanh: !b44
641[04:19:07.904] noobs_killer: Allie>Lco>SF!
642[04:19:08.082] Nightbot: B44 Sicilian Defence, 2.Nf3 e6 3.d4 cxd4 4.Nd4 Nc6 and B44 Sicilian, Szén Variation (5.Nb5)
643[04:19:17.328] tinduz_r: If there are vegas bets on allie making ot sufi, they must be feeling good right now.
644[04:19:43.374] atanh: @HenriKar_Chess Sure it's still zero. Just playing previous versions of itself instead of always the latest version.
645[04:19:56.342] NeelBasant: Next game will be interesting.
646[04:20:35.901] HenriKar_Chess: @atanh I'd think so, but who am I? :P
647[04:20:46.790] frakty_zero: but why should allie be better at beating weaker opponents? Is more due to network or search differences?
648[04:20:49.235] noobpwnftw: @sometimesithurts so I have replaced sizes with estimated number of positions for dbcn
649[04:20:57.393] zeekza: komodo playing the marotzcy bind interesting\
650[04:21:20.448] Jeroen_TCEC: @zeekza First 8 moves were book
651[04:21:22.488] noobpwnftw: note that it may be far off but the sizes seems to be off too
652[04:21:30.966] Bossieman: Leela 21,4% win rate S15 in divP. In div 16 Leela win rate is 18,2%
653[04:22:10.309] zeekza: i know but I mean like it's typically a pree positional opening so it'll be interesting to see komodo try to play it vs an NN
654[04:22:28.081] noobs_killer: !reasons
655[04:22:28.150] atanh: !cdb
656[04:22:29.346] chessdbcn (9b) • b6 -41 ! (25-03) • pv b6 Be3 Bb7 a3 h6 f4 O-O Qd3 • <ChessDBCN>
657[04:22:55.233] jigglefart: wow allie playing unreal
658[04:23:22.276] zeekza: a hedgehog, also typically a long manuevering positional type structure
659[04:23:23.949] Jeroen_TCEC: This is going to be a hedgehog type position.
660[04:23:49.348] noobs_killer: Reasons Leela is not performing 1. SSS, 2. Opening is drawish 3. not best net 4. GPU not doing well
661[04:23:52.724] zeekza: it already is a hedgehog isn't it
662[04:24:23.208] zeekza: wdym gpu not doing well? is there smtn wrong with the gpu
663[04:24:35.450] gakuj505: PogChamp PogChamp PogChamp
664[04:24:44.290] Jeroen_TCEC: It started as a Sicilian. In the hedgehog black's Nc6 is normally on d7
665[04:25:29.691] frakty_zero: yes they turn off GPu fans whenever leela plays
666[04:26:19.091] tinduz_r: There's something oddly beautiful with Black's structure.
667[04:26:22.547] dmuk77: If SF wins few more games quickly(which it should) then LC0 will be out of SuFi
668[04:26:38.799] atanh: !db
669[04:26:40.286] eval_bot (10b) +2=6-5 • Bb7 (+0=3-4) • O-O (+2=3-1) • [1995] Yermolinsky, Alex (2560) vs De Firmian, Nick E (2605): draw • [1993] Mokry, Karel (2525) vs Horvath, Jozsef (2535): draw • <Lichess>
670[04:26:41.175] HenriKar_Chess: @frakty_zero you forgot your Kappa there
671[04:27:12.489] dmuk77: Unless SF crashes again
672[04:28:20.063] dmuk77: And if the SuFi is between SF-Allie, SF will win this again
673[04:28:24.366] frakty_zero: @zeekza ok jokes aside but really in S14 I think there really was a GPU overheating oroblem that affected leela performance
674[04:29:10.932] zeekza: was lc the only nn in S14?
675[04:29:35.710] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
676[04:29:38.321] lmabacus: Scorpio was also NN
677[04:30:08.800] noobs_killer: The chance of LC0 vs SF final is very slim now. It could be Allie vs Lco or Allie vs SF.
678[04:30:35.372] HenriKar_Chess: @zeekza google for 'leela chess burrito memes'
679[04:30:49.593] zeekza: lmao wut
680[04:30:53.734] dmuk77: Yup seems like Allie will make it to SuFi.
681[04:31:45.473] frakty_zero: @zeekza when the overheating problem happened leela was the only NN engine, but I am not 100% if it was S14 (slight chance it was S13 I don't renenber)
682[04:31:53.273] noobs_killer: SF making Sufi will be like <5% now, 1. Low score 2. high risk of crush
683[04:32:20.238] ladaova: or crash
684[04:32:33.195] zeekza: leela chess burrito memes not finding much
685[04:32:47.504] lmabacus: I think S13. Don't think she would've made it to the Sufi with that problem.
686[04:32:48.559] dmuk77: If SF doesn't crash again then it might make it to SuFi as LC0 is not winning much
687[04:33:26.370] HenriKar_Chess: well, the top hit mentions a interview with one of its devs
688[04:33:38.642] HenriKar_Chess: mentioning the overheating issues
689[04:33:45.492] dmuk77: SF can still crush bottom 3 engines
690[04:34:13.440] zeekza: oh yea ok I was looking at the videos and images that came up
691[04:34:15.615] gakuj505: sf was yesterday
692[04:34:31.540] lmabacus: SF has no black wins surprisingly
693[04:35:12.286] zeekza: there are 2 black wins not from allie and 3 from allie
694[04:35:45.874] zeekza: I think that says a lot
695[04:35:57.048] ladaova: !92
696[04:35:57.307] lmabacus: And all but one off of KMCTS
697[04:36:12.391] atanh: Wow, this game is pretty crucial for Komodo, according to @Aloril42 's simulations. If it wins, it will have 24% chance making it to SuFi (based only on the game results provided to the simulation). If it loses, tho, that drops to 12%. A draw would leave it with 17% chance. Pretty big swing.
698[04:36:22.367] lmabacus: !time 92
699[04:36:23.578] eval_bot Game #92 "Stockfish 190826" vs "LCZero v0.22.0-nT40B.4-160" is estimated to start on Sat, 14 Sep 2019 11:18:31 GMT (After 2h 41')
700[04:37:57.660] lmabacus: Two straight Crashfish watches after this
701[04:38:07.389] zeekza: ok so 2/3 of the allie black wins are vs kmcts for 2/2 of the non allie black wins are also vs kmcts
702[04:38:16.105] jigglefart: it's crazy how amazing allie is doing here but is not even allowed over at banana
703[04:38:44.391] lmabacus: I thought gonzo refused to have it participate
704[04:39:04.338] zeekza: that means that allie is the only engine to have a win as black not vs kmcts
705[04:39:12.877] atanh: Game 99: Elsie vs. AllieStein. <hey that rhymes!> That'll be and interesting one.
706[04:39:22.971] theo77186Fr: I'm somewhat rooting for a 3rd SF crash
707[04:39:27.594] saurus1969: So Allie is claiming a place in the final, not relaying on gift
708[04:39:28.377] lmabacus: Yeah, against Houdini. Most impressive stuff
709[04:39:41.414] theo77186Fr: like I have ever rooted for any engine that crashed twice
710[04:40:16.497] zeekza: also tho all the black wins are from NNs \
711[04:40:47.027] zeekza: I'm still rooting for stoof
712[04:42:02.898] saurus1969: @theo77186Fr it is likely you will get the 3rd fish crashing. Emotions will increase if it comes late in the tournament DarkMode
713[04:42:26.305] saulzar: @jigglefart Allie is not allowed there yet Leelenstein is?
714[04:42:37.520] jigglefart: that is correct
715[04:42:57.311] saulzar: What's the logic there?
716[04:43:02.821] zeekza: allowed where?
717[04:43:05.296] ladaova: what banana ???
718[04:43:10.747] theo77186Fr: SF likelihood to crash again: 73.2%
719[04:43:22.033] Stephane_Nicolet: better late than never
720[04:43:23.468] lmabacus: They both were allowed at one point
721[04:43:23.473] jigglefart: c c c
722[04:43:25.219] theo77186Fr: (1/x improper prior)
723[04:43:43.720] theo77186Fr: and 85.6% with uniform prior
724[04:43:57.297] theo77186Fr: (not taking account of test games)
725[04:44:35.232] jigglefart: it appears that allie is clearly better than sf now. how is that even possible?
726[04:44:51.980] saulzar: There must be some reasoning... even if it is unfair?
727[04:44:54.770] gakuj505: PogChamp PogChamp PogChamp
728[04:45:25.324] lmabacus: Well, roughly equal in tournament play, infinitely better at not crashing.
729[04:45:26.116] zeekza: what are you talking abut?
730[04:45:43.399] Jeroen_TCEC: Guys, keep faith. Perhaps there will be no 3rd SF crash.
731[04:46:05.477] saulzar: Allie vs SF final
732[04:46:18.076] atanh: @noobpwnftw Do the estimations get better over time, or would it be more like a consistent error depending on the state/activity of the system at the time? It might be interesting to track some of the estimates over time to see how much they fluctuate and/or drift around. :) BTW, it's really cool. Currently estimated at 7,824,113,789 positions! PogChamp
733[04:47:12.686] theo77186Fr: Even with test games at 1/3 of a game, with the more favorable prior (1/x), the likelihood of crash is still 69.3%
734[04:47:58.954] theo77186Fr: So crash still likely
735[04:47:59.015] atanh: !faketemp
736[04:47:59.777] Nightbot: The current GPU temperature is 47 degrees Celsius KappaWealth
737[04:48:29.763] atanh: !cdb
738[04:48:30.907] chessdbcn (14w) • Qb3 66 ! (01-00) • pv Qb3 Nd7 • <ChessDBCN>
739[04:48:38.071] lmabacus: !chattemp
740[04:48:39.019] Nightbot: The current CHAT temperature is 494 Kelvin KappaWealth
741[04:48:44.510] Nuclearball: Did KMC donate another win to Allie?
742[04:49:52.426] gakuj505: ResidentSleeper
743[04:52:14.279] Jeroen_TCEC: !8ball No more crashes for SF in Div P?
744[04:52:14.894] Nightbot: All signs point to yes...
745[04:53:04.562] lmabacus: Those ellipses seem ominous
746[04:55:00.583] fresh_salsa_chips: stockfish's horoscope: You may find yourself winning but mysterious crashes will prevent you from getting what you really want Kappa
747[04:55:47.784] Martian_the_Martian: Yes, no more crashes.
748[04:57:00.397] theo77186Fr: With a likelihood of a SF 3rd crash >50% we can't rule out definitively this senario
749[04:57:02.625] Matuiss3: It would be funny to see SF crushing L1
750[04:57:06.242] theo77186Fr: *scenario
751[04:57:45.127] atanh: @fresh_salsa_chips Stockfish: "Mysterious crashes? Geez, I was gonna go grocery shopping today in the SUV, but maybe I should just stay home." Kappa
752[04:57:47.419] noobpwnftw: @atanh the estimations are from rocksdb, which due to the nature of file format it cannot be 100% accurate. I believe it is more accurate than size because it relies on some historical/statistical data
753[04:58:21.055] FoxMulder4: Even if SF won't crash anymore, other engines going clown-gaming against Allie gave a lot of point to her, I really can see either Leela (more likely) or SF missing SuFi.
754[04:58:26.882] Martian_the_Martian: @atanh use a bike, that is better.
755[04:59:24.795] theo77186Fr: The #1 in this L1 scored 67.9% in that. Guess the SF score if it crashes again LUL
756[04:59:29.406] atanh: Is it like Microsoft time estimates, though? 7.9 billion positions... 4.6 billion positions ... 100 billion positions... Kappa
757[04:59:38.902] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
758[04:59:42.034] chessinprogress: Allie! Allie!!
759[04:59:56.785] d0rus: Allie is on fire :o
760[05:00:01.467] Hanamuke: regardless, allie is on roll this season, it would be tied with sf woithout the crashes
761[05:00:13.847] atanh: @Martian_the_Martian Bike crashes can be more fatal than car crashes. SF doesn't have a helmet. BibleThump
762[05:00:22.088] chessinprogress: I knew Allie was going to get into the mix for this Sufi.
763[05:00:36.654] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
764[05:00:53.477] DanceWithMeQt: damn Allie is on steroids
765[05:01:42.263] chessinprogress: Leela’s problem is that she does not beat the weaker engines enough.
766[05:02:05.509] lmabacus: Solution: beat SF three more times
767[05:02:16.227] atanh: !cdb
768[05:02:17.532] chessdbcn (17w) • g3 86 ! (01-00) • pv g3 Bf6 • <ChessDBCN>
769[05:03:06.841] chessinprogress: What are the odds of SF crashing once more?
770[05:03:21.524] lmabacus: !sfprob
771[05:03:21.786] Nightbot: see https://i.imgur.com/HUYlRze.png by lmabacus
772[05:03:26.059] Jeydra: @chessinprogress quite good, the division only just passed the hardway mark
773[05:03:44.339] FoxMulder4: Would be great to see kind of playoffs format in DivP with top-4 engines. I am almost sure Leela can demontage Allie in a long series.
774[05:03:48.883] Martian_the_Martian: Striking that without SF crashing, there is a high probability that LCZero doesn't qualify for the SuperFinal.
775[05:03:56.550] Jeydra: *halfway mark
776[05:04:00.891] Dabbler1: wow Allie won again
777[05:04:20.545] FoxMulder4: @Martian_the_Martian This possibility is pretty much still there.
778[05:04:31.506] Martian_the_Martian: Yes
779[05:04:41.418] chessinprogress: @Martian_the_Martian yes...we had that discussion here over a month ago after that bonus where Allie came second and Leela third.
780[05:05:03.309] sometimesithurts: !arsenal
781[05:05:48.732] Martian_the_Martian: If SF survives today, playing 2 games in a row, the probability of SF not crashing a 3rd time, have increase significantly I think.
782[05:06:15.794] ladaova: Allie simply ROCKS
783[05:06:26.556] lmabacus: If you mean the rapid bonus, they tied for second and Leela won the head-to-head tiebreaker.
784[05:07:03.485] chessinprogress: @lmabacus what is that probability curve based on.
785[05:07:26.838] chessinprogress: @lmabacus in that rapid bonus, I thought Allie came 2nd.
786[05:08:14.012] lmabacus: Chances of SF crashing is 2/(number of games played), which is our best guess right now
787[05:08:15.129] Hanamuke: This leela need some contempt 4Head
788[05:08:49.660] lmabacus: Chance in each game, that is
789[05:09:25.192] chessinprogress: Ok...ty
790[05:09:25.207] kkeronen62: Fish still alive here, fine.
791[05:09:28.205] FoxMulder4: @Hanamuke was considered a meme but turned out to be reallity.
792[05:10:03.783] Martian_the_Martian: And then there is a.o. the fast rise of Stoofvlees, with its interesting play. The fight for the SuperFinal has become very competitive and interesting again.
793[05:10:40.485] lmabacus: It's a shame that Stoof and Scorpio didn't get net updates
794[05:11:23.610] FoxMulder4: @lmabacus where to find those rapid games?
795[05:11:52.106] lmabacus: Let me find the link
796[05:12:07.835] atanh: !time 99
797[05:12:09.104] eval_bot Game #99 "AllieStein v0.5-dev_1359f44-n10" vs "LCZero v0.22.0-nT40B.4-160" is estimated to start on Sun, 15 Sep 2019 08:24:32 GMT (After 23h 11')
798[05:12:33.469] Occyroexanthub: https://www.tcec-chess.com/archive.html?season=15&div=rapid&game=1
799[05:12:39.600] Occyroexanthub: @FoxMulder4 ^^
800[05:12:42.838] HenriKar_Chess: @FoxMulder4 you want the archive website or the PGN?
801[05:12:56.157] HenriKar_Chess: the PGN download link is also there ^^
802[05:13:07.878] chessinprogress: @lmabacus you are correct...Leela came second in the rapid bonus on tie break.
803[05:13:13.412] FoxMulder4: @Occyroexanthub great job, thanks!
804[05:14:18.885] lmabacus: All right, going to nap for a couple hours and hopefully catch the big game
805[05:18:21.328] skembris: I am not allowed to display the games that has been played. Anyone knows why?
806[05:19:28.888] chessinprogress: @lmabacus that will be a reverse of theFrench that Leela won againstSF.
807[05:19:47.777] Occyroexanthub: @skembris https://tcec-chess.com/
808[05:19:51.160] Occyroexanthub: are you here?
809[05:20:03.070] chessinprogress: Do you guys think that SF will beat Leela in that reverse French?
810[05:20:37.643] Occyroexanthub: maybe
811[05:20:43.223] Occyroexanthub: not that useful answer :P
812[05:20:55.186] fish_stock: re you on this site https://tcec-chess.com/
813[05:20:56.583] Occyroexanthub: i even believe Leela has chances to win
814[05:21:06.134] Occyroexanthub: i give 30-60-10
815[05:21:42.602] atanh: !cdb
816[05:21:43.930] chessdbcn (21w) • cxb5 29 ! (05-02) • c5 22 ! (01-01) • pv cxb5 axb5 Bxb5 Rxc1 • <ChessDBCN>
817[05:21:47.895] Occyroexanthub: but if i judge from CCCC recent results of SF-Leela in French i would say 60-39-1 for the result
818[05:22:40.773] ladaova: SF Probability of Crash is 1/(#remaining games)
819[05:23:11.992] Occyroexanthub: so it will crash in last game? :P
820[05:23:31.411] Occyroexanthub: *exactly before the last game
821[05:24:25.318] ladaova: BEcause the Immortal S16 must be achieved. SF gotta be out anway :-)
822[05:25:57.727] atanh: !cdb
823[05:25:58.941] chessdbcn (22w) • Qxc1 31 ! (05-01) • pv Qxc1 axb5 Bxb5 e5 f5 Ne7 a4 Qb8 • <ChessDBCN>
824[05:26:46.478] skembris: thanks - the link worked for me
825[05:26:48.423] noobpwnftw: all your positions belong to noob Kappa
826[05:27:50.781] ladaova: Belonga noob :-P
827[05:29:21.222] Jeroen_TCEC: Let's hope no more SF crashes. I still want to see a real battle for 2nd place!
828[05:29:36.885] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
829[05:30:47.443] bohan99: Has SF had a fix applied to stop the crashes?
830[05:31:42.646] Hanamuke: no @bohan99 someone vetoed it
831[05:32:03.131] chessinprogress: @Occyroexanthub I think SF has been playing the white side of the French well these days.
832[05:33:07.352] ladaova: SF becomes obsolete. Sooner or later
833[05:33:50.593] lunarzale: !blueleela
834[05:33:50.781] Nightbot: tcecLC v0.22.0-dev-nJ13B.2-136 (24x320), 2080 Ti and 7-piece tcecTB on SSD
835[05:34:07.047] jhl10: this has been a really interesting season so far, not expecting allie to even be in the conversation for the SUFI but is now firmly in 1st, SF crashing twice and leela struggling to win games
836[05:34:07.426] ndg2: Allie on the rise. How could she surpass her older cousin so fast?
837[05:34:19.686] Occyroexanthub: SF at +3.33 on this fortress PogChamp https://i.postimg.cc/2rW5wFW3/fhfhfh7777.png
838[05:34:44.548] Occyroexanthub: after 1 billion shuffling moves Sf on +3.23 and LEela at +0.08
839[05:34:49.739] Occyroexanthub: NotLikeThis
840[05:35:22.950] enqwert81: Allie is more aggressive than leela.
841[05:36:12.112] Bossieman: Is Allie AB in endgames yet? Is it stil lthe plan?
842[05:37:02.462] Martian_the_Martian: So, what caused SF to crash twice, is it fully certain and understood?
843[05:38:10.401] Occyroexanthub: they have some strong speculation
844[05:38:22.855] Occyroexanthub: no proof of course. and it's not easy to have a proof anyway
845[05:39:16.484] Martian_the_Martian: I heare all the time the dll is to blame, but how?
846[05:40:29.054] nsousa254: DLLs are EVIL
847[05:41:05.620] ladaova: WHy is not TCEC on Linux anyway ?
848[05:41:08.649] nsousa254: And they scam da fish!!!
849[05:41:51.860] Occyroexanthub: because many engines do not support Linux
850[05:42:49.444] saulzar: I just realised div-P reached the 50% NN based engines. How long until 100%? Knocking SF out would take a lot
851[05:43:00.398] theo77186Fr: You know how to disrupt TCEC: set TCEC on 128c/256th on Linux and enable HT
852[05:43:13.308] theo77186Fr: Boom. Non-native engine stuck on 64c.
853[05:43:36.510] theo77186Fr: (unless we fire a KVM instance + ssh)
854[05:44:20.406] theo77186Fr: Serve engine authors right PunOko
855[05:44:26.546] jim_vs: !51
856[05:44:26.703] Nightbot: In absence of further evidence, and as the logs didn't provide counter evidence, TCEC has no choice but to rule this as a tcecSF crash. See also https://github.com/official-stockfish/Stockfish/issues/2291 , https://github.com/official-stockfish/Stockfish/issues/2229 , and the Crash info tab - see also !74 and !dll
857[05:44:43.821] atanh: !cdb
858[05:44:45.080] chessdbcn (28w) • Bf2 86 ! (05-00) • pv Bf2 Nxf2 Rxf2 d5 b5 Qd6 • <ChessDBCN>
859[05:45:15.773] Occyroexanthub: a-b passers and yet draw?
860[05:45:18.707] theo77186Fr: And even KVM doesn't transmit CPU topology afaik
861[05:45:23.572] Occyroexanthub: interesting
862[05:45:50.749] kanchess: opp color passers = draw tcecK
863[05:47:12.070] kanchess: !threat lizard vlees
864[05:47:12.266] Nightbot: lizard is no threat to a modern, fine tuned, majestic engine with as much raw sex appeal as vlees tcecK
865[05:47:24.998] noobpwnftw: @theo77186Fr to serve them right we should boot into EFI shell and run engines from there, sorry no OS available you have to make your own Kappa
866[05:48:10.184] theo77186Fr: LUL
867[05:48:10.700] saurus1969: I would expect white to be much better is this position AsianGlow
868[05:48:53.024] zeekza: the king is pree open tho so
869[05:49:06.016] ladaova: other color kings = draw
870[05:49:24.113] zeekza: LUL Kappa
871[05:49:46.000] noobpwnftw: when I get my hands on some supercomputer they told me yeah well we use a patched gcc3.x and everything
872[05:49:51.635] kanchess: whoever thought of and created cdbcn is a genius tcecK
873[05:50:16.379] ladaova: cdbcn ??????????
874[05:51:04.513] kanchess: GIYF tcecK
875[05:51:55.774] kanchess: and the new special command is brilliant too
876[05:51:55.824] noobpwnftw: the theory has been there for a very long time but people had no faith until I actually build it
877[05:52:10.801] kanchess: ? x 1000
878[05:52:55.337] noobpwnftw: told you it'll replace RF
879[05:52:57.685] noobpwnftw: Kappa
880[05:53:03.042] kanchess: indeed
881[05:53:20.297] kanchess: how would a live vision of it look in GUI i wonder
882[05:53:39.738] sirinial64: Allie really wants to get her first sufi ever!
883[05:53:42.927] kanchess: caould be done in a different way perhaps than a "mere kibitzer" tcecK
884[05:54:01.261] saurus1969: @ladaova you are right, as long, as there are kings of different colors on the board, that are not mated, it is a draw. How could I miss that StrawBeary
885[05:54:16.526] noobpwnftw: I believe live can be reduced to JS SF + db probing
886[05:54:34.718] noobpwnftw: whenever there is a db result, show it, otherwise run localfish
887[05:54:47.980] kanchess: right
888[05:54:58.048] alayant: DansGame
889[05:55:13.211] atanh: !cdb
890[05:55:14.399] chessdbcn (31w) • Qb2 84 ! (12-00) • pv Qb2 Qe3 Bf1 dxe4 a6 • <ChessDBCN>
891[05:55:18.194] skembris: anyone knows if fat fritz will join the circus?
892[05:55:35.703] kanchess: what circus
893[05:55:37.411] alayant: It's a private leela clone
894[05:55:38.530] noobpwnftw: come on it refines the graph over time, it is much more accurate than it is now
895[05:56:31.174] skembris: are private engines not allowed?
896[05:56:42.485] saulzar: clones aren't
897[05:56:58.532] sirinial64: So weird. On tcec-chess I cannot connect with my twitch account in the twitch chat window. But it is working fine from the twitch website
898[05:57:19.871] gnocchipup: test
899[05:58:05.600] gnocchipup: I'm using chrome, tcec site chat works for me (I can't do it on the android phone chrome though)
900[05:58:18.035] sparkleofdiego: whoa, Allie-Stein in the lead! PogChamp
901[05:58:39.304] skembris: yes Leela got lazy
902[05:58:49.520] Occyroexanthub: in the comfortable lead actually
903[05:58:51.672] Occyroexanthub: for now :P
904[05:58:55.938] gnocchipup: but leela has 0 losses, so far
905[05:59:10.049] skembris: and Stockfish Rusty
906[05:59:22.318] hackerbot101: Okay I'm new here. What is Allie-Stein? Is it new?
907[05:59:22.724] saurus1969: Are there any fat Fritz test with reasonable sample size?
908[05:59:28.632] Occyroexanthub: tough schedule ahead for allie though
909[05:59:32.736] gnocchipup: how is allie different from leela?
910[05:59:38.643] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
911[05:59:46.330] Occyroexanthub: Stoof-Allie, Allie-Leela, SF-Allie !
912[06:00:04.050] ladaova: Whats the next game please ??!!
913[06:00:16.702] Occyroexanthub: SF-Leela
914[06:00:17.339] skembris: check schedule
915[06:00:35.841] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
916[06:00:41.226] ladaova: The Immortal game !
917[06:00:51.648] alayant: Allie : https://cdn.discordapp.com/attachments/561184936418213889/621995751970439171/unknown.png
918[06:01:52.318] gnocchipup: oh - found the allie github now, hehe
919[06:02:00.958] jimthree60: oh boyo, AllieStein is threatening to keep fish out of the sufi seriously :O
920[06:02:27.736] alayant: Either SF crashes a 3rd time and is out
921[06:02:30.603] alayant: Or SF is in SuFi
922[06:02:41.731] alayant: And Leela is out
923[06:03:08.021] flitzekacke: OR Leela is in Sufi either way and SF is out LUL
924[06:03:10.470] jim_vs: Leela reproducing season 10 SF? Kappa
925[06:03:12.316] aartdappels: well no
926[06:03:41.214] Vizvezdenec: well truth to be told allie blue-black cloud is now more blue
927[06:03:50.802] Vizvezdenec: because of ideas it took from ethereal LUL
928[06:04:18.124] aartdappels: (god, please get rid of the 30 seconds slow mode!!) leela is equal with stockfish don't know why stockfish has the advantage
929[06:04:40.218] sometimesithurts: so should it be blue-black-red LUL
930[06:04:42.583] alayant: Claiming the training script used by the stein net was 100% the one from Lc0 is also incorrect afaict, jjosh did experiments with cyclical LR and other differences
931[06:04:53.063] tcecrob: SB
932[06:04:56.207] alayant: But the picture gives a good general idea
933[06:05:20.514] alayant: (Also SL needs some different training methods compared to RL anyway)
934[06:05:32.689] gnocchipup: @alayant thnaks
935[06:05:59.495] Vizvezdenec: stein is like 90% of leela net 30 games
936[06:06:15.233] Vizvezdenec: so more or less the same leela that played in season 14
937[06:06:24.377] Vizvezdenec: it's like... Taking eval from sf8
938[06:06:27.164] alayant: For the source games it uses a lot of leela training games
939[06:06:33.693] Vizvezdenec: giving it sf dev search
940[06:06:38.778] Vizvezdenec: will this be unique?
941[06:06:48.685] Vizvezdenec: "a lot" is 90%
942[06:06:51.855] Vizvezdenec: or even more
943[06:06:54.834] jim_vs: Komodo is not so far. Only one loss.
944[06:07:24.691] ladaova: Who let ALlie in TCEC ?
945[06:07:48.253] Occyroexanthub: 90 % is not right
946[06:08:12.579] aartdappels: I really dislike the decision to use T40 games, would not accept if I was TCEC
947[06:08:16.481] Occyroexanthub: if i recall Stein net is CCRL+other sources 20 % games, and 80% T30 games
948[06:08:42.619] aartdappels: even 80% T30 games seems pretty ridiculous to me
949[06:08:44.053] Occyroexanthub: but i recall that Jjosh had said in June he started getting T40 games also
950[06:09:00.493] NikolayKralev: next game predicitions ?
951[06:09:15.222] Occyroexanthub: Jjosh also said he will add 10 million T40 games to Stein net but will ask Anton if it's ok with it
952[06:09:18.118] Occyroexanthub: i find it ok though
953[06:09:27.049] Occyroexanthub: games should be public knowledge
954[06:09:37.716] Occyroexanthub: only GAMEs though
955[06:09:41.101] Occyroexanthub: not training code etc
956[06:09:47.547] saurus1969: Also CCRL games are derived from other engines....
957[06:09:48.576] Occyroexanthub: for TCEC restrictions i mean
958[06:10:10.997] aartdappels: yeah public knowledge, not necessarily a separate TEC entry. same training code used and for now roughly the same search algorithm. alltogether I find the two too similar
959[06:10:11.529] alayant: If a NN used fishtest games as a training source, I wouldn't care at all
960[06:10:35.915] sometimesithurts: you mean AF Kappa
961[06:10:47.420] alayant: (It happens that due to adjudication, a game saving bug, and other elements, these are not a good training source)
962[06:10:48.714] aartdappels: @alayant a completely different situation obviously, since stockfish doesn't train on fishtest games
963[06:11:56.015] Vizvezdenec: http://tests.stockfishchess.org/tests/view/5d7cb9c40ebc5902d3864b62 this test doesn't fail in 500 games LUL
964[06:11:58.816] alayant: It's different in that the games generated by the leela project are with a different purpose than those generated at fishtest, but... If you assume the leela training games should be a kind of propriety of the leela project that can't be used by other projects, then shouldn't the same be said of fishtest games ?
965[06:12:02.442] aartdappels: if stein nets are moving towords leela nets and not away from them with non-zero sources then I just don't know what the point of it is
966[06:12:32.089] Vizvezdenec: point is to earn money
967[06:12:39.351] aartdappels: @alayant my problem with it is that when you use the same training data you just get a net with the same play style and biases. training data is crucial for the identity of a net
968[06:12:44.371] Occyroexanthub: games are too be shared and it should not be nobody's property
969[06:12:46.427] Vizvezdenec: since author sells them
970[06:12:52.221] Occyroexanthub: LEela's or fishtest's
971[06:13:11.106] Vizvezdenec: point of allie is to boost your ego via basically cloning yourself into finals of tournaments
972[06:13:31.309] aartdappels: everybody can use leelas data, I 'm just saying that I don't see the point of doing so for a separate TCEC entry
973[06:13:47.341] alayant: @aartdappels Well, I'm not arguing in favor of having nets that play the same way
974[06:13:47.476] Occyroexanthub: it's a different engine obviously
975[06:14:12.653] Occyroexanthub: of course that doesn't mean it doe snot shared/used many things from LEela
976[06:14:49.734] aartdappels: god I hate this 30 second slow mode, for context, 10M T40 games is almost everything, LD2 (the cpu leela net) only used like 8M T40 games, and was trained with the last 3M games for the most part
977[06:14:51.604] alayant: Anyway current TCEC rules about how NN uniqueness sux, but even with different rules that are not awful, there are both many shared and many different things with Leela in Allie.
978[06:14:54.819] Occyroexanthub: shared LEela's code for some things in binary/search/backend/training code but it's a distinct engine now!
979[06:14:56.897] Stephane_Nicolet: @alayant Any update on the 3-ply book idea?
980[06:14:57.382] naw72: what if someone set up a training network of their own, with self play learning, would it end up looking similar to leela's games now? what if the training method was similar, but not same?
981[06:15:02.413] saulzar: Vizvezdenec Allie is a clone as much as A/B engines are all clones of each other...
982[06:15:07.521] Aloril42: @alayant Not sure if it's there or missing in that diagram, scoring endgame positions using tcecSF search with 7-piece Syzygy and using those in training of Leelenstein.
983[06:15:21.581] Occyroexanthub: if TCEC calls it different fine by me
984[06:15:28.563] alayant: @Stephane_Nicolet 6-ply book rather. I talked a bit with noob about it yesterday but he hasn't gone around generating it yet.
985[06:15:34.154] Occyroexanthub: but TCEC's rules for NNs have to improve
986[06:16:06.525] Stephane_Nicolet: thanks
987[06:16:07.955] alayant: @Aloril42 Difference in training code/method that the diagram misses
988[06:16:09.781] aartdappels: in the end I would find a leela - allie Sufi just very boring because of the similarities at the moment
989[06:16:11.289] saulzar: Allie has almost the same algorithm as Lc0 but then Houdini, Komodo and SF have pretty much the same algorithm, too
990[06:16:36.231] ladaova: from Rybka
991[06:16:41.934] saulzar: Looks like Allie vs SF if SF can somehow not crash
992[06:16:45.450] GamBito93: @aartdappels: is LD2 really the stronger NN for CPU?
993[06:17:08.194] aartdappels: @GamBito93 yeah, at least short to medium TC
994[06:17:21.794] Occyroexanthub: Allie-LEela sufi would be snoozefest
995[06:17:28.103] oci82: SF bug still there?
996[06:17:28.132] toxikcake: bye bye SF, hope you come back nest tcec
997[06:17:31.358] Aloril42: !nocrash I hope tcecSF doesn't crash, that should keep exitement until end.
998[06:17:31.613] Nightbot: galacticAWW galacticDERP galacticDERP galacticDERP
999[06:17:35.215] Occyroexanthub: Allie-Sf or Leela-Sf of course would be much better
1000[06:17:45.344] Occyroexanthub: i prefer Leela <3
1001[06:17:50.233] Stephane_Nicolet: ah, I was wondering how many seconds after loging in without "crash" in the chat
1002[06:18:14.604] aartdappels: I'd rather see Komodo vs Houdini than leela allie lol, it would be like T40 vs T30 playing
1003[06:18:32.500] Occyroexanthub: Komodo Houdini would be even more boring
1004[06:18:40.935] saulzar: Leela vs SF seems very difficult from here, Allie has a decent lead ..
1005[06:18:44.490] Occyroexanthub: been there. we have seen that
1006[06:18:57.676] aartdappels: yeah SF-leela is almost impossible now
1007[06:18:59.644] Aloril42: @aartdappels Oh, @jhorthos says there is now stronger CPU net.
1008[06:19:28.148] aartdappels: @Aloril42 which one, there are a few candidates
1009[06:19:30.929] Occyroexanthub: 19-20 games left
1010[06:19:33.886] Occyroexanthub: so nothing impossible
1011[06:19:43.408] saulzar: Only unlikely
1012[06:19:53.607] Occyroexanthub: not even unlikely
1013[06:19:57.222] Occyroexanthub: with 19 games left
1014[06:20:12.199] alayant: Definitely unlikely
1015[06:20:15.206] saulzar: Would require SF to not crash and Leela to win 4 more games than Allie, given that the opposite has occurred so far
1016[06:20:18.618] Aloril42: Have to ask him, forgot to do it.
1017[06:20:23.856] Martian_the_Martian: Judgement Day!
1018[06:20:26.174] alayant: Probability of SF crashing in the last 20 games is significant
1019[06:20:27.247] sirinial64: I really want to see if leela will hold the reverse #92 against fish
1020[06:20:31.747] Stephane_Nicolet: This guy is a genius: https://www.youtube.com/watch?v=lAFnftbMQmA
1021[06:20:40.736] aartdappels: its highly improbable leela vs SF will play because allie has a 2 point lead to both atm and the chances of stockfish crashing are substantial too
1022[06:20:41.474] Occyroexanthub: well ok if SF crashes it's obviously unlikely
1023[06:20:44.837] Aloril42: (script) 91: 4. Komodo 3808(-3) - 6. Stoofvlees 3760(+1) tcecDD https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-91-komodo-2381-00-stoofvlees-ii-a12/
1024[06:20:45.784] Occyroexanthub: but i think it will be fine
1025[06:20:55.741] alayant: Then if it doesn't crash, the probability of Allie coming behind both Leela and SF is small
1026[06:20:59.831] Occyroexanthub: not 2 points. 2 points with 1 game more
1027[06:21:05.385] Occyroexanthub: probably 1.5 points after next game
1028[06:21:07.362] sirinial64: Because of crashes, sf loses every tiebreaks
1029[06:21:07.771] ladaova: THe next game is the - Saturday Night Fever
1030[06:21:14.078] aartdappels: yeah but a leela-sf game, so 1.5 avg
1031[06:21:19.263] alayant: For a SF-Leela final, you don't need just SF or Leela ahead of Allie, but both ahead of it
1032[06:21:42.177] frakty_zero: oh here we go reverse of glorious leela win
1033[06:21:50.907] mrbdzz: Leela game!
1034[06:21:51.869] Aloril42: (script) if 1-0: Sto: 3842(+7), tcecLC 3832(-6); if tcecDD : Sto: 3834(-1), tcecLC 3839(+1); if 0-1: Sto: 3828(-7), tcecLC 3845(+7); simulation and estimated start times: https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-92-speculation-simulation/
1035[06:21:58.268] alayant: Without the crashes, Allie would have 0.5 point less and SF 1 more, so it would be much more likely
1036[06:22:01.170] alayant: !hypefish
1037[06:22:01.407] Nightbot: fl0mHear ? ? ? ? ? ? renayHype
1038[06:22:05.075] alayant: !prayfish
1039[06:22:05.313] Nightbot: BlessRNG stay stable MercyWing1 SabaPing MercyWing2 stay stable BlessRNG
1040[06:22:05.995] hackerbot101: Why is Stockfish crashing so much?
1041[06:22:07.659] saulzar: Yep. Allie would have to tank
1042[06:22:09.239] sirinial64: !hype this game !
1043[06:22:09.456] Nightbot: aweSome
1044[06:22:09.549] mrbdzz: !goelsie
1045[06:22:09.750] Nightbot: laceFight
1046[06:22:39.717] sirinial64: !gostable crashfish !
1047[06:22:39.833] Free3ze: hold baby BabyRage
1048[06:22:39.922] Nightbot: galacticLUCK galacticLURK galacticLUCK
1049[06:22:47.996] Stephane_Nicolet: Stockfish is not crashing, the pthread library is :-)
1050[06:23:04.475] alayant: @hackerbot101 Bug in MSYS2 Mingw's winpthread code. Compiler issue.
1051[06:23:23.497] Stephane_Nicolet: it's my fault
1052[06:23:25.694] Martian_the_Martian: :O
1053[06:23:28.125] ladaova: If not a mistake of SF then why is not DQ the DLL ?!
1054[06:23:29.961] saulzar: @Stephane_Nicolet Windows specific??
1055[06:23:40.863] sometimesithurts: no it's Cato's
1056[06:23:55.736] mrbdzz: !reverse
1057[06:23:55.903] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
1058[06:23:59.417] eval_bot Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1059[06:24:07.370] saulzar: Who runs any real software on windows these days anyway ;)
1060[06:24:10.729] Aloril42: @alayant I can update dll or exe (same version for exe) for CPU bonus.
1061[06:24:13.373] atanh: The game that just ended is a potential false draw from the drawrule. I suspect one of the two engines would have blundered. There were different lines showing white at advantage and black at advantage. I think it was more of a dynamic balance than a dead draw. Hard to say for sure, but could be considered/checked-out if the false draw concept becomes a real issue.
1062[06:24:14.027] mrbdzz: Diverged already: 0-1
1063[06:24:15.609] Occyroexanthub: French a scam for favoring SF as it has white Kappa
1064[06:24:22.250] alayant: You asked for the optimized build, I submitted it, nobody caught it used the same buggy compiler as CoffeeOne in july...
1065[06:24:22.252] frakty_zero: 0-0 - the wrong plan
1066[06:24:24.876] Stephane_Nicolet: @saulzar yes, even specific to a specific compiled version for Windows
1067[06:24:38.224] saulzar: Ouch :(
1068[06:24:46.801] Occyroexanthub: 0-0 is ok
1069[06:24:55.935] Occyroexanthub: white cannot win only with h4 etc
1070[06:24:58.692] Aloril42: @alayant Just 14 games at 30min+5s for tcecSF and total time a bit over 2 days.
1071[06:24:59.367] ladaova: HERE WE COME !!!!!
1072[06:25:03.441] mrbdzz: So it's buggy compiler, not buggy pthread lib?
1073[06:25:09.943] sirinial64: !reverse leela not going b5-b4
1074[06:25:10.162] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
1075[06:25:13.044] eval_bot [10:25:42] Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1076[06:25:16.542] frakty_zero: @alayant was buggy compiler really the cause? :o
1077[06:25:24.365] Aloril42: @alayant So not much more than testing 10 games at same TC.
1078[06:25:29.750] NikolayKralev: draw soon
1079[06:25:34.272] alayant: @frakty_zero Yes.
1080[06:25:50.116] alayant: @Aloril42 What's the easiest for you ?
1081[06:26:06.219] NikolayKralev: LC0 is the best drawmaster so
1082[06:26:18.795] Aloril42: @alayant Either one is OK as long tcecSF version stays same.
1083[06:26:20.252] frakty_zero: @alayant how is that possible,?
1084[06:26:33.847] Aloril42: @alayant Maybe you want dll fix?
1085[06:26:35.959] josephviruses: Oh, the reverse of Leela's win!
1086[06:26:38.494] kanchess: !gostable gooooooooo SabaPing
1087[06:26:38.653] Nightbot: galacticLUCK galacticLURK galacticLUCK
1088[06:26:50.146] josephviruses: !goelsie
1089[06:26:50.364] Nightbot: laceFight
1090[06:26:53.765] NikolayKralev: if draw than LC0 will pass
1091[06:26:55.042] kanchess: !c05
1092[06:26:55.280] Nightbot: C05 French, Tarrasch, Closed Variation
1093[06:27:03.697] sirinial64: !prayfish very good command
1094[06:27:03.887] Nightbot: BlessRNG stay stable MercyWing1 SabaPing MercyWing2 stay stable BlessRNG
1095[06:27:05.811] josephviruses: !reverse
1096[06:27:05.964] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
1097[06:27:09.115] eval_bot [10:27:38] Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1098[06:27:25.196] Aloril42: !next @alayant No hurry, there is plenty of time until end of DivP
1099[06:27:25.443] Nightbot: Next DivP CPU engines + top 4 CPU engines from League 1 Bonus (Laser, tcecLC CPU, Ethereal and Xiphos), 30min+5s, 1x DRR, a bit over 2 days. Deadline for Sufi is last move of Bonus. Then testing for Sufi. Then Sufi.
1100[06:27:35.353] sirinial64: !cdb
1101[06:27:36.710] chessdbcn (10w) • Qe1 51 ! (01-00) • pv Qe1 Be7 f4 O-O Nf3 f5 exf6 Nxf6 • <ChessDBCN>
1102[06:27:48.201] Stephane_Nicolet: @mrbdzz As I understand it, an update of the compiler was shipped with a buggy pthread library, so I don't know how to call it axactly
1103[06:27:52.419] fishtrawler: Go SF!
1104[06:27:52.675] frakty_zero: who has better chances to win this game? (taking into consideration that SF crashed only wit black so far)
1105[06:27:57.368] mrbdzz: right, it's not buggy compiler per se (i.e., gcc/clang generating wrong code); it's buggy support lib
1106[06:27:58.678] Shlapfish: Go crashfish!!!
1107[06:28:00.037] kanchess: do i see a new badge Aloril42 <3 nice!!!
1108[06:28:04.546] josephviruses: SF diverged immediately from the get go
1109[06:28:10.799] kanchess: deep purple
1110[06:28:11.955] mrbdzz: but that support lib is shipped with the compiler suit
1111[06:28:15.737] mrbdzz: suite
1112[06:28:31.905] kanchess: !db
1113[06:28:32.986] eval_bot (10b) +0=0-0 • No moves • No games • <Lichess>
1114[06:28:37.015] kanchess: !cdb
1115[06:28:37.731] chessdbcn (10b) • Be7 -22 ! (01-01) • pv Be7 f4 Nc7 Nf3 O-O a4 g6 Be3 • <ChessDBCN>
1116[06:28:58.730] Stephane_Nicolet: yes
1117[06:28:59.859] Martian_the_Martian: Hmm, could one argue that it wasn't SF that crashed?
1118[06:29:06.931] sirinial64: We are in for a treat
1119[06:29:36.493] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
1120[06:29:44.118] sometimesithurts: It seems a draw LUL
1121[06:29:44.620] frakty_zero: @mrbdzz and this support lib, in what way was it buggy, mismatched versions or simply corrupted binary?
1122[06:29:45.121] Aloril42: @Martian_the_Martian That was basis for voting to fix dll.
1123[06:29:47.941] Stephane_Nicolet: anyway, the situation is what is it and we have to live with it :-)
1124[06:30:06.286] alen5199: !boom
1125[06:30:06.442] Moobot: the concept of the "boom" is a TCEC invention [a BOOM means "eval explodes away from 0.00" - the opposite is a MOOB which therefore means "eval implodes back in the general direction of 0.00"]
1126[06:30:08.590] Occyroexanthub: it's crazy Nb8 is actually a move NotLikeThis
1127[06:30:25.191] mrbdzz: @frakty_zero Just functional bug in the library. The library is pthread support code for windows; and there's a bug that causes hangs
1128[06:30:26.254] alayant: !gosf
1129[06:30:26.458] Nightbot: zizWhale
1130[06:30:44.924] alayant: The bug cause deadlocks
1131[06:30:51.326] mrbdzz: ^
1132[06:31:16.923] frakty_zero: @mrbdzz that really sux :/
1133[06:31:17.645] Martian_the_Martian: Maybe TCEC admins have quietly replaced the buggy dll, who could tell...
1134[06:31:35.732] GamBito93: !boom5
1135[06:31:35.922] Nightbot: diamondFire diamondFire diamondFire diamondFire diamondFire bueRocket
1136[06:31:36.690] Jeydra: !gofish
1137[06:31:36.925] Nightbot: bungieSOFFISH
1138[06:31:40.163] Lc0fan: is prob. of crash constant each move or dependent on game state?
1139[06:31:49.277] ladaova: SF +1.40 !
1140[06:31:50.181] mrbdzz: @frakty_zero Well, there's really no bug-free software. :/
1141[06:32:04.219] mrbdzz: SW is written by hoomans, after all
1142[06:32:07.484] Jeydra: Let's have SF play to a M score and then crash
1143[06:32:18.296] alayant: @Lc0fan We don't know !
1144[06:32:23.578] sometimesithurts: except for sf code, which is bug free :)
1145[06:32:26.442] Lc0fan: thanks
1146[06:32:31.607] alayant: Unlikely that it's constant for each move I'd say
1147[06:32:32.085] Occyroexanthub: Leela wants immediate Nc6 NotLikeThis
1148[06:32:37.818] mrbdzz: @sometimesithurts Right :)
1149[06:32:42.951] Occyroexanthub: blueLeela did not like that
1150[06:32:46.463] NikhilChess: sometimes backward moves are so dangerous
1151[06:33:14.588] alayant: This DivP is getting more and more exciting LUL
1152[06:33:16.744] fresh_salsa_chips: backwards moves in the french Kappa
1153[06:33:19.957] Occyroexanthub: redSF definitely did not like Nc6
1154[06:33:21.583] kkeronen62: Nc6 N at 365chess
1155[06:33:21.795] Aloril42: Could be constant for time spent searching? Unless it's particular type of position that triggers it.
1156[06:33:25.672] Martian_the_Martian: Does SF crash randomly, or is it more likely under certain circumstances?
1157[06:33:43.882] frakty_zero: is leela aiming at counyrtplay on d4 ? maybe even get Qeen to b6 later on (just speculating
1158[06:33:47.390] Occyroexanthub: hmm redSf got tricked again :P
1159[06:33:47.601] Nightbot: athHM
1160[06:33:59.927] sometimesithurts: @Occyroexanthub isn't black leela more powerful than Blue one?
1161[06:34:04.007] atanh: Already out of book? Sheesh. Looks like the French needs more research.
1162[06:34:08.476] HenriKar_Chess: If it's a threading bug, I hazard the guess that it's more likely running with more threads
1163[06:34:12.301] alayant: !cdb
1164[06:34:13.708] chessdbcn (12b) • exf5 -95 ? (00-00) • <ChessDBCN>
1165[06:34:22.458] Occyroexanthub: black Leela has 2 GPUs but blueLeela has better net
1166[06:34:33.935] HenriKar_Chess: this can of library is specifically there to _avoid_ deadlocks and hangs in multi-threaded programs
1167[06:34:36.981] mrbdzz: blueleela has also ponder
1168[06:34:39.320] HenriKar_Chess: can -> kind
1169[06:34:42.328] Aloril42: Blueleela gives eval for white moves too.
1170[06:34:43.650] sirinial64: !blueleela
1171[06:34:43.859] Nightbot: tcecLC v0.22.0-dev-nJ13B.2-136 (24x320), 2080 Ti and 7-piece tcecTB on SSD
1172[06:34:44.928] sometimesithurts: better net? it uses J13?
1173[06:34:50.599] Occyroexanthub: yes
1174[06:35:01.959] HenriKar_Chess: I guess there may be a bug in one of the implementations of deadlock avoidance
1175[06:35:06.857] alayant: The fishes really hate black's position
1176[06:35:09.291] Occyroexanthub: wow redSf thinks black is already a toast?
1177[06:35:20.810] frakty_zero: @Occyroexanthub it's most likelt better net only on strong HW
1178[06:35:32.992] Occyroexanthub: CCCC has shown Sf knows to play this French with white and LEela does not!
1179[06:35:33.361] ladaova: @Occyroexanthub Why the Lc0 does not use the best net here?
1180[06:35:33.586] kanchess: nice vid Stephane_Nicolet great reallies
1181[06:35:35.695] kkeronen62: 9. - Rc8 was already nearly never played
1182[06:35:38.934] Occyroexanthub: so this might be real 1-0
1183[06:35:39.920] alayant: Redfish is very unstable
1184[06:35:51.140] spg1990: sf should not crash for 3rd time . I want to see sf vs lc0
1185[06:35:53.181] HenriKar_Chess: !redfish
1186[06:35:53.388] Nightbot: tcecSF 2019090303 (C=0) running on 192C box with 256 threads in VM limited to 256 vcores and 6-piece tcecTB . Around 4x speed and +50 ELO compared to the 43 core Sufi tcecSF . Hardware: 8xXeon Scalable 8168 192C/384T. Part of !arsenal
1187[06:36:00.354] alayant: !sf1-0
1188[06:36:00.501] Nightbot: SF has 105 win cheers bungieSOFFISH
1189[06:36:07.502] oci82: draw
1190[06:36:08.208] essedaly: what is whitefish contempt?
1191[06:36:08.522] mrbdzz: !leela0-1 tcecK
1192[06:36:08.749] Nightbot: A total of 348 wins have been cheered for Leela tcecK
1193[06:36:08.815] HenriKar_Chess: hmm.. C=0 => contempt set for 0 ?
1194[06:36:26.900] alayant: Yes C is shorthand for contempt
1195[06:36:29.037] kanchess: !sf1-0
1196[06:36:29.265] Nightbot: SF has 106 win cheers bungieSOFFISH
1197[06:36:35.093] mrbdzz: cmonBruh
1198[06:36:41.917] frakty_zero: @ladaova I think it's because none knew how it will scale to TCEC hw, at cccc they use more powerful HW
1199[06:36:43.390] sometimesithurts: they implemented the Ostrich algorithm for deadlock Kappa
1200[06:36:48.995] kanchess: !l-eela tcecK
1201[06:36:49.297] gakuj505: Kappa
1202[06:36:58.745] mrbdzz: @kanchess Trolling?
1203[06:37:01.022] mrbdzz: tcecK
1204[06:37:01.239] atanh: Bluela basically has 'ponder' on, so should give deeper, more stable PVs.
1205[06:37:05.578] kanchess: :D tcecK
1206[06:37:10.488] essedaly: @alayant what is contempt of the stockfish playing?
1207[06:37:11.994] HenriKar_Chess: @alayant is there an explanation for redfish being 'very unstable' ?
1208[06:37:12.094] ma_d____m_an: @spg1990 if SF didn't crash the SUFI will be allie vs SF
1209[06:37:12.790] mrbdzz: :|
1210[06:37:15.358] mrbdzz: LUL
1211[06:37:17.359] kanchess: not trolling, just teasing
1212[06:37:20.245] gakuj505: 1 point for both :))
1213[06:37:25.186] alayant: @essedaly 24
1214[06:37:41.148] essedaly: thanks @alayant
1215[06:37:42.318] oci82: @ma_d____m_an fake news
1216[06:37:47.780] kanchess: if i woud catch myself trolling i would ban myself - too dangerous tcecK
1217[06:37:59.634] mrbdzz: lol
1218[06:38:24.080] alayant: @HenriKar_Chess In complex non-equal positions, AB evals often swing significantly as the PV changes a lot and so do the evals of the leaves node.
1219[06:38:37.408] sometimesithurts: wow trolling leads to ban? so what is tcec chat for all about? LUL @kanchess
1220[06:38:56.050] L0RINDA: I used to troll a lot and they punished me by making me mod
1221[06:38:56.883] mikebridges: !reverse
1222[06:38:57.073] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
1223[06:39:00.109] jim_vs: !reverse
1224[06:39:00.241] eval_bot Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1225[06:39:00.407] eval_bot: [10:39:29] "!reverse" will be available in 11.8 sec.
1226[06:39:03.252] HenriKar_Chess: @alayant OK, But that's specific for Redfish, right? SF should also have this, plus the contempt
1227[06:39:17.037] kanchess: the only way to beat L0RINDA was to join L0RINDA
1228[06:39:22.967] L0RINDA: :D
1229[06:39:23.044] frakty_zero: imho leela more likely to win this game than SF (if you know what I mean):D
1230[06:39:27.046] mikebridges: !db
1231[06:39:28.106] eval_bot (14b) +0=0-0 • No moves • No games • <Lichess>
1232[06:39:32.808] sometimesithurts: @L0RINDA that seems toooo harsh for such an act
1233[06:39:39.434] nickpelling: @L0RINDA making the punishment fit the crime? The bastards.
1234[06:39:40.923] alayant: The contempt changes the preferred lines slightly, but it shouldn't reduce or increase much the swinging tendencies
1235[06:39:41.793] fresh_salsa_chips: rook on f5 PogChamp
1236[06:39:45.184] ma_d____m_an: @oci82 not really 100% fake ;)
1237[06:39:45.345] L0RINDA: @sometimesithurts They used me as an example
1238[06:39:45.436] HenriKar_Chess: specific -> not specific
1239[06:39:53.562] mrbdzz: Same thing happened to me, @L0RINDA DansGame
1240[06:39:57.389] mikebridges: big plus from the opening for white, sf needs to make it count
1241[06:40:18.938] sometimesithurts: I'll stop trolling :( I swear
1242[06:40:22.232] oci82: so 2 crashes of 22 games? 9% chance of crash
1243[06:40:23.652] gakuj505: ResidentSleeper
1244[06:40:24.266] nickpelling: 18. Rg3 was basically the end of it in the reverse
1245[06:40:29.406] mrbdzz: @sometimesithurts LOL
1246[06:40:39.250] georgelvoinicel: If SF crashed twice in 74 games what is the probability to crash once in 168-74=94 games?
1247[06:40:55.769] positiev3xxx: @georgelvoinicel they fixed it
1248[06:40:55.811] Nuclearball: Leela doesn't understand king danger so I think if she castles this will escalate quickly
1249[06:41:00.157] HenriKar_Chess: @oci82 91% chance of no crash in 1 game . 81% chance of no crash in 2 games ... ... ...
1250[06:41:02.311] Martian_the_Martian: SF probability of crashing is 2/42.
1251[06:41:10.294] gakuj505: leela PogChamp and fish PogChamp
1252[06:41:11.172] frakty_zero: does anyone know if apparently allie being better than leela at beating weaker opponents is due to different net or different search?
1253[06:41:17.161] Stephane_Nicolet: !crashprob
1254[06:41:17.363] Nightbot: In the Middle Ages they were trying to calculate how many angels you can put on a pin head. These crashing probability calculations remind me a little bit of that. --Stephane_Nicolet
1255[06:41:19.814] L0RINDA: The probability behind it all is rather complex and slightly philosophical too
1256[06:41:29.887] georgelvoinicel: @positiev3xxx how if were not allowed?
1257[06:41:32.500] oci82: @HenriKar_Chess yep
1258[06:41:47.683] gakuj505: El Clásico Kappa Kappa Kappa KappaClaus
1259[06:41:58.509] HenriKar_Chess: !cdb
1260[06:41:58.792] atanh: Funny, SF evaled the opening at around 0, now it started at around 0.8. Is contempt set to 40?
1261[06:41:59.590] chessdbcn (14b) • O-O -94 ? (01-01) • cxd4 -119 ? (01-01) • Rc7 -128 ? (01-01) • Ra8 -144 ? (01-01) • a6 -148 ? (01-01) • g6 -176 ? (01-01) • pv O-O Nf1 • <ChessDBCN>
1262[06:42:00.937] positiev3xxx: @georgelvoinicel they got permission
1263[06:42:02.005] Lc0fan: prob of crash is zero because sf has never crashed 3 times
1264[06:42:04.406] mrbdzz: fishmoobs!
1265[06:42:05.560] L0RINDA: They largely overestimate the chance of a crash though
1266[06:42:17.696] Noobilow: The only way now to make SF lose is making it crash LUL
1267[06:42:19.414] L0RINDA: uh, that is that a large number of the estimates are a little high
1268[06:42:37.039] L0RINDA: as, i suspect, were the people calculating them
1269[06:42:37.181] kanchess: !hypesf SabaPing
1270[06:42:37.401] Nightbot: cobaltL SabaPing cobaltR
1271[06:42:37.724] simbaaaaaaaa: probability of SF crashing again is 50% - it will either happen or it won't and 50% is the middle
1272[06:42:41.505] georgelvoinicel: @positiev3xxx afaik the voting said no
1273[06:42:53.417] Nuclearball: As long as SF avoids shuffling there will be no crash
1274[06:42:59.367] positiev3xxx: @georgelvoinicel Really? I thought it was mostly yes?
1275[06:43:00.375] nickpelling: The best way to avoid a crash is to win early and hard. Luckily SF is good at that. :-)
1276[06:43:05.440] sometimesithurts: there should be a "crash" word counter in chat LUL
1277[06:43:11.671] fish_stock: no crash this game, however the next game 88%
1278[06:43:18.116] HenriKar_Chess: !do
1279[06:43:18.349] Nightbot: The TCEC-should-do-this-and-that Counter is now at 591 nadedKing
1280[06:43:19.131] gakuj505: Nf8 PogChamp PogChamp PogChamp
1281[06:43:30.348] oci82: @simbaaaaaaaa fake news
1282[06:43:30.544] Nuclearball: Put crash as a curse word and chat would be empty LUL
1283[06:43:31.182] positiev3xxx: !vote
1284[06:43:31.251] nickpelling: @positiev3xxx one no meant all no
1285[06:43:31.316] L0RINDA: @positiev3xxx it needed to be a 100% vote as it was to override the set rules
1286[06:43:31.365] Nightbot: ""Should Stockfish team be allowed to replace libwinpthread-1.dll with non-buggy version?" AntonMihailov: The final results of the voting are now in. Out of 8 votes we have 6 "yes", 1 "abstain", and one "no". This final no is the veto. Voting is anonymous and TCEC team will respect that and the outcome of the authors' vote nadedKing
1287[06:43:33.985] NikolayKralev: this game is draw
1288[06:43:35.959] sometimesithurts: It will overflow in a day
1289[06:43:44.476] georgelvoinicel: @positiev3xxx houdart voted against and this was a veto
1290[06:43:46.011] Noobilow: This patch is totaly fine...not a single crash in my machine after thousands of games.
1291[06:43:50.357] Martian_the_Martian: That Knight wants to go to e6.
1292[06:43:53.231] kanchess: !threat PunOko SabaPing
1293[06:43:53.405] Nightbot: PunOko is no threat to a modern, fine tuned, majestic engine with as much raw sex appeal as SabaPing tcecK
1294[06:44:01.387] sirinial64: Nf8-Ne6 protects g7 and draw Kappa
1295[06:44:05.319] positiev3xxx: I didn't know, okay
1296[06:44:05.967] L0RINDA: We dont know who voted against as it was an anonymous vote (unless I missed some news)
1297[06:44:23.697] kanchess: it's anonymous
1298[06:44:29.302] georgelvoinicel: @L0RINDA yeah
1299[06:44:29.762] kanchess: and it will remain that way
1300[06:44:37.162] alayant: Leela team held a public vote
1301[06:44:45.673] positiev3xxx: then it is obviously bigger than %50, get ready for relegation of SF
1302[06:44:45.800] gakuj505: PogChamp PogChamp
1303[06:44:46.919] enqwert81: Nf8 is interesting
1304[06:44:47.756] snipertooth: !reverse
1305[06:44:47.951] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
1306[06:44:48.679] HenriKar_Chess: well, SF has, on second thought, been convinced by Nf6
1307[06:44:51.384] eval_bot Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1308[06:45:03.306] frakty_zero: @L0RINDA so let's speculate who might that have been xD
1309[06:45:11.025] Nuclearball: !reverse
1310[06:45:11.095] L0RINDA: !speculate
1311[06:45:11.285] Nightbot: Speculation about the SF Vote is allowed (if not encouraged), but please do not ask the devs, even in jest, who they voted for
1312[06:45:13.864] eval_bot [10:45:43] Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1313[06:45:32.302] kanchess: ^^ wtg L0RINDA
1314[06:45:32.985] atanh: Instamove already
1315[06:45:36.024] Stephane_Nicolet: draw :(
1316[06:45:36.337] Martian_the_Martian: N is very nice on e6.
1317[06:45:40.815] NikolayKralev: first game was 1-0 for LC0 or this is new ?
1318[06:45:42.327] HenriKar_Chess: ehm I meant Nf8
1319[06:45:47.780] gakuj505: 0.00 s PogChamp PogChamp PogChamp
1320[06:45:54.829] enqwert81: Blue leela didn't like Nf8.
1321[06:45:57.587] sometimesithurts: TCEC should announce "Houdart is the NO vote".. to avenge him for not replying with an update for H
1322[06:45:59.146] jim_vs: Rxf7?
1323[06:46:03.700] chmariach: draw... or crash?
1324[06:46:03.828] alayant: I don't think this is a draw.
1325[06:46:41.098] sometimesithurts: !moob
1326[06:46:41.344] positiev3xxx: It was obviously either Allie or Leela team
1327[06:46:43.819] kanchess: you don't know that sometimesithurts how can you not announce what you do not know huh tell me
1328[06:46:45.411] Nuclearball: The side with a pawn protected on e5 has much higher chances to win
1329[06:46:47.703] fresh_salsa_chips: that f file PogChamp
1330[06:46:49.726] oci82: clearly draw
1331[06:46:56.098] L0RINDA: Houdart is killed by the villagers. "You fools, I was the Seer" he cries. Night falls. During the night you hear fighting and screaming. After dawn breaks, the villagers find the corpse of Allie. Anton takes a new vote. 4 yes, 1 abstain, 1 no.
1332[06:46:58.305] georgelvoinicel: Some devs maade public their vote, and some smart guys found out by elimination that Houdard vote against.
1333[06:47:13.945] sirinial64: (No-yes-abstain) is (Stoof-Houdini-Komodo) but the correct order is unknown
1334[06:47:22.354] sometimesithurts: @kanchess the point is to avenge.. we don't want to hear the truth Kappa
1335[06:47:28.790] mrbdzz: I thought Gonzo already publicly announced that he voted 'NO' tcecK
1336[06:48:00.902] mrbdzz: he also said publicly he voted 'YES'... tcecK
1337[06:48:03.723] xoto10: Ah, the witch-hunt is still on I see LUL
1338[06:48:08.193] chmariach: you can't make your vote public; you can just make a public claim regarding what you voted
1339[06:48:20.466] nickpelling: witches the truth?
1340[06:48:20.627] xoto10: Is someone witchfinder general?
1341[06:48:28.019] sirinial64: @L0RINDA the village wakes up LUL
1342[06:48:34.570] L0RINDA: ;)
1343[06:48:40.886] xoto10: @chmariach exactly :)
1344[06:49:03.180] xoto10: bit late for this game, any notable moves in the opening?
1345[06:49:04.658] snipertooth: this looks dicey
1346[06:49:05.461] fresh_salsa_chips: could the black king have a safe home on d7? KappaPride
1347[06:49:16.192] Nuclearball: D@x
1348[06:49:23.047] sometimesithurts: @xoto10 yeah Nb8 Kappa
1349[06:49:23.596] Nuclearball: No h4
1350[06:49:35.013] georgelvoinicel: No witch hunting. Just figuring out who voted against.
1351[06:49:37.969] Noobilow: They will crash SF again...it can win
1352[06:49:51.534] mrbdzz: which hunting tcecK
1353[06:50:00.073] enqwert81: There is no safe place for black king.
1354[06:50:11.437] mrbdzz: oops
1355[06:50:11.468] chmariach: so... Good Witch Hunting?
1356[06:50:18.460] GamBito93: !boom5
1357[06:50:18.693] Nightbot: diamondFire diamondFire diamondFire diamondFire diamondFire bueRocket
1358[06:50:32.866] alayant: WTF is this boom, arent Ncxd4 and Nexd4 equivalent by transposition ?
1359[06:50:33.058] Nightbot: What The Fish nukeKarp
1360[06:50:33.070] sirinial64: After "the king in the north" "the king in the open board" Kappa
1361[06:50:34.053] sometimesithurts: OML
1362[06:50:34.263] Nightbot: Oh My Leela PunOko
1363[06:50:40.284] fresh_salsa_chips: !fake
1364[06:50:40.467] Nightbot: attempts to spread fake news may get you banned
1365[06:50:44.802] flashyduckling: Which hunting?
1366[06:50:52.698] alayant: Immediate Rxg7 !!
1367[06:50:55.992] enqwert81: Leela can lose this.
1368[06:50:59.064] Martian_the_Martian: Fireworks on the board now.
1369[06:51:00.949] alayant: Go go
1370[06:51:10.510] fresh_salsa_chips: !fakeboom
1371[06:51:10.700] Nightbot: F clintusRAID A clintusRAID K clintusRAID E
1372[06:51:10.903] georgelvoinicel: Anway I find the veto rule as counter productive. People want to see the two best engines in sufi. I dont care about Houdart or Gonzo or whoever vote.
1373[06:51:20.956] Nuclearball: Leela trying to out-tactic SF LUL
1374[06:51:29.608] alayant: Redfish now wants Ng3...
1375[06:51:30.953] HenriKar_Chess: @xoto10 14... Nf8 was a big surprise
1376[06:51:31.748] Martian_the_Martian: I think game will become more tactical now.
1377[06:51:34.427] positiev3xxx: We need more opening favouing one side just like this game. This is more fun.
1378[06:51:45.172] alayant: This is getting really interesting
1379[06:51:50.078] fish_stock: who on earth would have voted no... ?
1380[06:51:51.463] HenriKar_Chess: 11.... Nc6 also, but less than Nf8
1381[06:51:54.724] alayant: Good time for a long fish think
1382[06:52:18.381] frakty_zero: SF could no longer handle pressure against d4
1383[06:52:19.139] hiep_vu: think too long and people freak out
1384[06:52:27.692] enqwert81: 1-0
1385[06:52:34.018] flashyduckling: No vote for what?
1386[06:52:43.681] josephviruses: Leela will hold this
1387[06:52:44.282] sometimesithurts: !vote
1388[06:52:44.488] Nightbot: ""Should Stockfish team be allowed to replace libwinpthread-1.dll with non-buggy version?" AntonMihailov: The final results of the voting are now in. Out of 8 votes we have 6 "yes", 1 "abstain", and one "no". This final no is the veto. Voting is anonymous and TCEC team will respect that and the outcome of the authors' vote nadedKing
1389[06:52:57.602] Bossieman: !db
1390[06:52:58.893] eval_bot (18w) +0=0-0 • No moves • No games • <Lichess>
1391[06:53:05.161] chmariach: unbooming slowly
1392[06:53:06.370] oci82: 0-1 crash
1393[06:53:07.002] atanh: It's a dead draw, no doubt. ResidentSleeper SF started his attack half-cocked and Leela will recover but fail to launch an attack of her own.
1394[06:53:09.142] Bossieman: !reverse
1395[06:53:09.302] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
1396[06:53:12.450] eval_bot Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1397[06:53:24.927] Nuclearball: This game now entered on forced long tactical win zone SF likes the most
1398[06:53:27.401] Bossieman: Is this a 1-0 1-0 opening?!?
1399[06:53:31.621] HenriKar_Chess: !db 2rqkb1r/p2n1ppp/1pn1p3/2ppP3/3P1P2/P1P5/1P1NN1PP/R1BQ1RK1 w k - 1 12
1400[06:53:32.017] eval_bot +0=0-0 • No moves • No games • <Lichess>
1401[06:53:38.251] HenriKar_Chess: this was Nc6
1402[06:53:48.079] hiep_vu: moob ??
1403[06:53:51.572] sometimesithurts: !divergance
1404[06:53:56.658] L0RINDA: @georgelvoinicel The entrants in the event have a right to have it played by the rules. I think nearly everyone would like a SF superfinal (if it earned it, which it probably would), but we're not the ones who put time into making an engine to play by a certain ruleset
1405[06:53:58.087] fresh_salsa_chips: black kingside vaporized Kappa
1406[06:54:21.209] positiev3xxx: Isn't black king in the middle of the battle zone? Looks scary
1407[06:54:27.822] josephviruses: Leela's King is now semi-exposed but she thi k's he is alright
1408[06:54:28.126] frakty_zero: @fresh_salsa_chips white cener as well though
1409[06:54:28.347] atanh: SF unwittingly opening files for Leela's upcoming attack on white king.
1410[06:54:29.667] fish_stock: by logical deduction it was either 1 or 8
1411[06:54:32.196] enqwert81: The position suits SF very much with black king in the middle.
1412[06:54:41.785] chmariach: !static
1413[06:54:43.030] eval_bot (18b) Total(5.28,1.40) • Material(0.74,0.41) • Imbalance(0.17,0.17) • Pawns(0.19,0.07) • Knights(0.03,0.04) • Bishops(-0.10,-0.11) • Rooks(0.11,0.23) • Queens(0.00,0.00) • Mobility(0.19,0.45) • King(4.09,0.60) • Threats(-0.48,-0.35) • Passed(0.22,-0.04) • Space(0.13,0.00) • Initiative(0.00,-0.06) <Stockfish>
1414[06:54:43.150] Nuclearball: Leela has 0 king danger. Not new
1415[06:54:48.517] CatoTheYounger_TCEC: this game started going sideways with 9...Rc8, a very rare move, and 10...Nab8 was a novelty
1416[06:54:58.476] Noobilow: "Forced long tactical win zone..." PogChamp
1417[06:55:04.669] L0RINDA: :o cato
1418[06:55:18.876] fish_stock: 0-0
1419[06:55:27.437] fresh_salsa_chips: leela TILTED Kappa
1420[06:55:32.999] Occyroexanthub: LEela with the stupid NExd4 NotLikeThis
1421[06:56:02.633] positiev3xxx: @Occyroexanthub What makes you think it's stupid :)
1422[06:56:08.797] oci82: @Occyroexanthub she seems fine according to evals what are you saying?
1423[06:56:09.562] Occyroexanthub: this is now black King in the center so a very probable 1-0
1424[06:56:11.813] mrbdzz: It's not clear to me that Nexd4 was too stupid
1425[06:56:21.608] georgelvoinicel: @L0RINDA SF was in sole lead. So I find the decision unfair and based on envy.
1426[06:56:21.797] josephviruses: Leela will be 1.5-0.5 with this opening
1427[06:56:22.294] Occyroexanthub: it allowed Rxg7
1428[06:56:38.290] chmariach: the decision to crash?
1429[06:56:41.004] alayant: Rxg7 was going to happen with a different move order
1430[06:56:42.864] enqwert81: Black king is vulnerable on light squares.
1431[06:56:45.565] Occyroexanthub: how black can even play with King in the center???
1432[06:56:49.589] atanh: !cdb
1433[06:56:50.818] chessdbcn (18b) • Rf8 -129 ? (01-01) • Qd7 -175 ? (01-01) • Ne6 -242 ? (01-01) • Rc7 -295 ? (01-01) • Nxe2+ -510 ? (01-01) • pv Rf8 Nf3 • <ChessDBCN>
1434[06:56:52.638] frakty_zero: with perfect play this should be draw however it's just this sort of position wher leela is much more likely to blunder than SF and one blunder here could mean game over
1435[06:56:56.529] Jeydra: Black's position looks so bad I'm surprised SF's eval is so low
1436[06:56:59.938] mrbdzz: Rf8 looks like a nice move to slow down the attack, and black's pieces are quite ok
1437[06:57:26.752] Jeydra: Like, king in center in open position, how is Black not getting blown off the board
1438[06:57:28.312] mrbdzz: the king and pawns are bad for black, though
1439[06:57:31.287] theo77186Fr: I'm waiting for a SF crash Kappa
1440[06:57:36.021] fresh_salsa_chips: bLacK pIEceS aRe qUiTe oK Kappa
1441[06:57:40.532] L0RINDA: @georgelvoinicel It's not unfair in any way. If you're saying it's a shame, I agree with you. Someone wanting the rules enforced is perfectly within their rights as a competitor though - there's also a case for saying it protects the future integrity of the tournament
1442[06:57:55.323] oci82: king in center to gobble up pawns
1443[06:58:00.389] Occyroexanthub: but i have to ask again. how black can even play with King in the center??? this seems so doomed for LEela
1444[06:58:04.419] theo77186Fr: !8ball will SF crash in these 2 games?
1445[06:58:04.983] Nightbot: Concentrate and ask again...
1446[06:58:13.781] Nuclearball: SF has a bad bishop, otherwise ok
1447[06:58:15.114] Jeydra: Exactly @Occyroexanthub not just king in center, Black's even down a pawn
1448[06:58:17.162] L0RINDA: concentrate pls theo
1449[06:58:20.472] atanh: @noobpwnftw There's something seriously wrong with CDBCN. It's showing Leela with lower than -100 eval. Must be a fatal bug.
1450[06:58:21.959] sometimesithurts: oh Nf3 Nxf3 .. you see Nxd4 not that stupid
1451[06:58:27.448] josephviruses: I guess leela has the mobility down pat Kappa
1452[06:58:28.769] Occyroexanthub: !9ball should TCEC ban the word crash in its chat?
1453[06:58:29.362] Nightbot: Ask Anton
1454[06:58:34.039] positiev3xxx: @Occyroexanthub Leela thinks she can push the attack back
1455[06:58:36.523] mrbdzz: black is also a bit ahead in overall development
1456[06:58:37.612] Occyroexanthub: damn NotLikeThis
1457[06:58:38.617] fresh_salsa_chips: @Occyroexanthub the king will march to the queenside, the pawns will move backwards, and leela will be safe KappaPride
1458[06:58:41.884] Occyroexanthub: nightbot is good
1459[06:58:53.469] L0RINDA: 8balll manages to make more sense than most of chat :D
1460[06:58:53.677] klaalb: gold answer
1461[06:58:56.562] gakuj505: boom draw !!! NotLikeThis
1462[06:58:58.904] NikolayKralev: seems LC0 is unbeatable
1463[06:59:07.860] L0RINDA: Need a NN 8ball
1464[06:59:22.925] frakty_zero: @Occyroexanthub because white pieces at the moment are not coordinated enough to take advantage of it - i.e. lack of development + black counerplay in the cener
1465[06:59:28.572] mrbdzz: Yes, let's spend 10000000 GPU hours to train 8nnball tcecK
1466[06:59:38.943] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
1467[06:59:39.703] georgelvoinicel: @L0RINDA I am definitely against the veto. This is my point. Why not majority as it is normal??
1468[06:59:41.091] josephviruses: I guess leela will be unbeatable again like in last divp. The question is will it land in sufi Kappa
1469[06:59:48.086] ma_d____m_an: @gakuj505 it is called moob
1470[07:00:08.818] Noobilow: !10ball should TCEC make SF crash again?
1471[07:00:13.787] Occyroexanthub: @georgelvoinicel because TCEC wanted to break their rules so ALL authors playing here should agree.
1472[07:00:16.045] klaalb: crash sf crash
1473[07:00:17.913] Occyroexanthub: do you understand it now?
1474[07:00:19.231] mrbdzz: right, so now everyone has bad pawns and a bit dubious king safety
1475[07:00:27.879] L0RINDA: @georgelvoinicel A majority would be used in deciding a new rule set for future events most likely (If it wasnt just a straight admin decision), to actively agree to not abide by the rules, it is fair to make sure there is not a single objection
1476[07:00:31.409] sometimesithurts: I will contribute if there will be a 8nnball
1477[07:00:36.002] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
1478[07:00:52.112] ladaova: LC0 negative soon !
1479[07:01:01.790] sometimesithurts: both by writing code and GPU time Kappa
1480[07:01:01.907] NikolayKralev: 0-1 soon
1481[07:01:12.766] Occyroexanthub: 0-3 even sooner
1482[07:01:15.084] Archstro: Where is the crash? We were promised a crash!
1483[07:01:18.210] theo77186Fr: 0-1 by crash that is LUL
1484[07:01:19.136] chmariach: 8nnball should run on the GPU machine and steal compute from the NN engines
1485[07:01:27.396] mrbdzz: cmonBruh
1486[07:01:55.682] theo77186Fr: And then throw both wins away because SF would have crashed thrice
1487[07:01:57.108] Occyroexanthub: if SF crashes Leela loses half a point compared to all other engines
1488[07:02:11.298] Nuclearball: SF with 2 passed pawns. Yeah 0-1 LUL
1489[07:02:18.514] hiep_vu: Allie too
1490[07:02:18.592] Occyroexanthub: +1.60 PogChamp
1491[07:02:20.692] positiev3xxx: @Occyroexanthub Not against Allie
1492[07:02:28.026] L0RINDA: @georgelvoinicel The ruling would have been that SF is not allowed to change. The engine devs were given a chance to allow that as a special case rather than have the admin ruling. If everyone agrees, then there is clearly no issue. If someone isn't happy, the pre-agreed rules should stand or it opens a whole can of worms
1493[07:02:34.586] Occyroexanthub: redSF just 0.62
1494[07:02:35.877] ma_d____m_an: If this game walk towards draw ...SF will have bigger chance of crashing as leela will intentionally drag the game to 400 moves
1495[07:02:35.908] NikolayKralev: mega moob
1496[07:02:37.141] Archstro: "Totally worth it to see SF relegate." --Leela
1497[07:02:37.715] mrbdzz: Qd7 nice move, temporarily protects h7
1498[07:02:50.566] sometimesithurts: sf crashes ==> leela loses half point ==> K-Allie sufi LUL
1499[07:02:58.087] alayant: SF admins didn't want to assume a decision of yes/no themselves
1500[07:03:05.427] gakuj505: SeemsGood love
1501[07:03:08.602] hiep_vu: Allie and H will lose .5 too
1502[07:03:11.647] mrbdzz: Take the free h7 pawn, fish! Take it!
1503[07:03:12.996] georgelvoinicel: I disagree. Giving a newbie like Scorpio, lets say, a veto right is not ok. Too much power for an insignificant engine. A normal vote is more than fair.
1504[07:03:22.489] oci82: why does leela lose half a point if sf crashes?
1505[07:03:27.622] sparkleofdiego: Komodo vs Allie-Stein SuFi! <3 PogChamp
1506[07:03:41.722] fresh_salsa_chips: div p engine is an "insignificant engine" Keepo
1507[07:03:50.365] hiep_vu: because SF would be disqualified and all matches with SF are removed
1508[07:03:52.421] L0RINDA: @georgelvoinicel Why is it not okay? All engines in the event have played by the rules to get this far. What if the question was "Let's kick out Scorpio for being new" surely Scorpio has earned the right to veto that
1509[07:03:58.530] gakuj505: boom pls NotLikeThis
1510[07:04:00.327] oci82: all points disregarded? ok
1511[07:04:03.955] mrbdzz: @oci82 All SF games will be removed, and Leela is +1 vs SF
1512[07:04:07.893] alayant: Scorpio has more right to veto than Houdini LUL
1513[07:04:07.951] positiev3xxx: @oci82 when an engine is relegated due to 3x crash, all of its matches are disgarded
1514[07:04:32.040] GamBito93: SF's eval is super unstable
1515[07:04:42.599] oci82: ok makes sense
1516[07:04:44.019] frakty_zero: @oci82 because she scored a win against SF earlier and after SF being disqualified all the SF games will not be counted
1517[07:04:47.810] Occyroexanthub: +1.55 PogChamp
1518[07:04:47.831] xoto10: sf sees pawn up ending but I'm not sure it's a win
1519[07:04:54.448] Occyroexanthub: strange as redSf is on +0.27
1520[07:04:56.439] Occyroexanthub: very strange
1521[07:05:00.283] L0RINDA: @georgelvoinicel Again, I can understand entirely why you're not happy with the result of the vote, but I cant understand why you think the process is unfair
1522[07:05:06.726] alayant: The divergence is on move 24
1523[07:05:07.160] Bossieman: 1.55 is very high for SF. Close to winning
1524[07:05:07.576] mrbdzz: That's contempt hallucination
1525[07:05:09.616] theo77186Fr: TCEC SF is seeing ghosts again
1526[07:05:13.894] alayant: Rg5 Ng6 vs Rg5 Nc6
1527[07:05:16.303] mrbdzz: RF is at 0.2
1528[07:05:21.051] Occyroexanthub: but redSF does not see this +1 Pawn up endgame?
1529[07:05:23.955] Occyroexanthub: it's still SF :P
1530[07:05:25.841] alayant: Leela has yet another line
1531[07:05:38.092] mariushav: Always interesting when there's a gap in evaluation
1532[07:05:42.735] L0RINDA: @georgelvoinicel Scorpio has been a part of this event for many seasons
1533[07:05:50.642] alayant: Whitefish's end of PV is a dead win
1534[07:05:55.582] Occyroexanthub: here the gap is between Sf and SF :P
1535[07:06:03.538] sparkleofdiego: RedFish at +0.23! why this huge eval discrepancy with SF, which is displaying +1.55!? can anyone please explain?? BegWan
1536[07:06:08.976] mariushav: Yup!
1537[07:06:11.338] georgelvoinicel: @L0RINDA Kicking out an engine is something I never heard of, but if thete is some evidence that the devs of an engine did something wrong why dont kick that engine out?
1538[07:06:11.397] alayant: But the issue is that it is likely missing some key black moves
1539[07:06:21.034] gakuj505: go go go
1540[07:06:22.366] Occyroexanthub: maybe redSf misses 25.Bb2 of SF?
1541[07:06:32.662] talkshowwcharles: Different dev builds
1542[07:06:45.593] L0RINDA: @georgelvoinicel There is evidence that SF devs did something wrong. Accidental, unlucky, and sad though that is
1543[07:06:51.049] alayant: Most likely is whitefish missing 24.... Ng6
1544[07:06:58.122] sometimesithurts: the gap in evals with rf and sf in agreement would have probably lead to 1-0 but this time I think not
1545[07:06:59.207] sirinial64: !redfish
1546[07:06:59.362] Nightbot: tcecSF 2019090303 (C=0) running on 192C box with 256 threads in VM limited to 256 vcores and 6-piece tcecTB . Around 4x speed and +50 ELO compared to the 43 core Sufi tcecSF . Hardware: 8xXeon Scalable 8168 192C/384T. Part of !arsenal
1547[07:07:05.278] Occyroexanthub: @alayant yes
1548[07:07:15.989] kanchess: georgelvoinicel made a typo there, he did something wrong! by his own words i can kick him out LUL tcecK
1549[07:07:17.451] d0rus: A veto in this situation seems prettu reasonable given everybody agreed on the standing rules before hand and rules where about to be changed mid game. That can only happen if all agree unanimous.
1550[07:07:20.930] Occyroexanthub: but maybe Leela miss it too :p
1551[07:07:25.908] alayant: If this is a win, this is 100% a contempt win...
1552[07:07:43.959] mrbdzz: oof... now everyone has to watch out for typos? monkaS
1553[07:07:54.273] georgelvoinicel: @L0RINDA then vote kicking SF out. Is there a majority?
1554[07:07:59.077] alayant: Well, 100% is not right but you get the point LUL Redfish is very pessimistic
1555[07:08:02.096] sometimesithurts: so then 0 contempt or 24 in sufi??
1556[07:08:03.159] mariushav: @alayant for white?
1557[07:08:04.491] GamBito93: Maybe SF is evaluating some draw endgame with more than +1.00, very likely
1558[07:08:05.678] L0RINDA: @georgelvoinicel Why is a majority fair, I don't understand
1559[07:08:06.006] positiev3xxx: @alayant What do you mean by contempt win? wouldn't be a win if C was 0?
1560[07:08:14.281] gakuj505: go leela
1561[07:08:17.092] xoto10: yeah, we handicapped our own engine, terrible behaviour
1562[07:08:17.430] Occyroexanthub: +1.73 PogChamp
1563[07:08:22.491] alayant: @positiev3xxx Yes, that's the idea
1564[07:08:24.553] L0RINDA: @georgelvoinicel The rule in place would have eliminated SF (Assuming a third crash). There was no need for a vote at all
1565[07:08:28.537] alayant: Whitefish isn't changing its mind
1566[07:08:34.352] GamBito93: !boom5
1567[07:08:34.538] Nightbot: diamondFire diamondFire diamondFire diamondFire diamondFire bueRocket
1568[07:08:38.662] alayant: And it has Ng6 now in PV !
1569[07:08:40.620] nickpelling: Sees Ng6 now
1570[07:08:43.789] d0rus: I dont believe SF +1.73 if redfish say 0.16 xD
1571[07:08:46.747] alayant: Still at sky-high eval
1572[07:08:49.840] frakty_zero: If only white fish sees it and others dont and if white fish turns out to be correct that will make whitefish the strongest engine on the planet[
1573[07:08:52.429] Archstro: Dreamfish in action
1574[07:08:55.200] alayant: Now this is getting very bizarre
1575[07:08:57.716] mrbdzz: I agree with @d0rus
1576[07:09:04.083] Occyroexanthub: whiteSF does not see 24...Ng6
1577[07:09:09.345] Occyroexanthub: but maybe Leela will not see it either
1578[07:09:22.851] alayant: @Occyroexanthub WhiteSF had 24... Ng6 in PV for a moment
1579[07:09:35.083] alayant: And still didn't moob. It went back to Nc6 at the end tho
1580[07:09:38.818] Jeydra: Wow @ the SF/RF divergence
1581[07:09:38.882] georgelvoinicel: Why you don't have the veto right in elections in your country? What if everyone has a veto right?
1582[07:09:39.471] xoto10: I suspect wf and rf are about equivalent, just different, when wf is playing and rf kibitzing. The kibitzing handicaps rf to an extent
1583[07:09:40.567] Occyroexanthub: this is gonna be interesting
1584[07:09:43.162] nickpelling: @Occyroexanthub it sees Ng6 (it was in the PV), but it obviously evaluates it differently.
1585[07:09:44.571] gakuj505: +1.8 PogChamp PogChamp PogChamp
1586[07:09:46.647] doppelbauer: oh, lunch is over ... now i need a dessert :)
1587[07:09:48.023] sparkleofdiego: +1.80!! from SF
1588[07:09:50.922] L0RINDA: @georgelvoinicel I live in the uk, they ignore votes here anyway
1589[07:09:56.320] alayant: Biggest divergence I ever remember between white and redfish
1590[07:09:59.383] Bossieman: Its over! 1-0
1591[07:10:17.455] josephviruses: Who to trust? Redfish is a noob
1592[07:10:19.460] doppelbauer: hmmm, only the white line ... this is not enough :)
1593[07:10:22.420] xoto10: lunch? I only just had breakfast :)
1594[07:10:24.035] Occyroexanthub: LEela will not see the defensive resource 24...Ng6! either :(
1595[07:10:25.191] theo77186Fr: Me thinks it's a ghost
1596[07:10:28.319] kanchess: an close in depth too, red and white
1597[07:10:31.922] josephviruses: Kappa
1598[07:10:32.697] L0RINDA: @georgelvoinicel But in most countries, the government is elected and then enforces and makes the rules. America doesn't have an election to find out what kind of toilet paper Trump should use
1599[07:10:36.407] Occyroexanthub: 1-0
1600[07:10:37.547] georgelvoinicel: @L0RINDA Brexit is on its way. Don't worry.
1601[07:10:40.243] fresh_salsa_chips: breakfast? I just had after dinner snack Kappa
1602[07:10:41.260] Occyroexanthub: if LEela does not see it
1603[07:10:42.191] xoto10: but yeah, this is dessert :)
1604[07:10:45.516] Aloril42: http://www.youtube.com/watch?v=u3L8vGMDYD8 "World's Best Table Tennis Robot vs TableTennisDaily's Dan!"
1605[07:10:51.814] doppelbauer: xoto austria :)
1606[07:10:54.523] sometimesithurts: hmm maybe a bonus here between rf and sf to see if rf can hold this
1607[07:10:56.882] alayant: Rg8+ ? Isn't the plan supposed to be Qg2 ? Whitefish ? :p :p
1608[07:10:57.965] theo77186Fr: I just ate lunch LUL
1609[07:11:00.561] Stephane_Nicolet: If it is a win, so much for the contempt=24 haters :-)
1610[07:11:13.138] Martian_the_Martian: :)
1611[07:11:13.766] positiev3xxx: Line diverges somewhere between depth 40 and 45 ? PogChamp
1612[07:11:33.298] chmariach: clearly this is a fortress
1613[07:11:40.910] josephviruses: Redfish maintains it's a draw. Leela too. Sup whitefish?
1614[07:11:43.596] Occyroexanthub: critical moment now
1615[07:11:49.484] bigdaas: ola amigos!
1616[07:11:51.688] theo77186Fr: moob
1617[07:11:53.228] Occyroexanthub: LEela has to show 24...Ng6 to survive!!
1618[07:12:04.103] d0rus: I wonder if whitefish will manage to blunder before it sees the draw.
1619[07:12:04.678] bigdaas: *hola
1620[07:12:08.497] L0RINDA: @georgelvoinicel Your reasoning is based on your presumption that a vote was something the devs were entitled to. It was not. The rules were already in place to deal with the situation.
1621[07:12:09.727] Occyroexanthub: if shw shows 24...Nc6 it's 1-0
1622[07:12:13.802] Occyroexanthub: *she
1623[07:12:30.050] NikolayKralev: 0.13 ?????
1624[07:12:34.973] bigdaas: so sf must not crash, else sf is out?!
1625[07:12:35.356] alayant: SF needs to finally play this Qg2 move first
1626[07:12:40.097] Martian_the_Martian: Still this 24. ... Nc6/Ng6 PV divergence between the fishes!
1627[07:12:44.349] Occyroexanthub: it will
1628[07:12:46.124] d0rus: @bigdaas yes
1629[07:12:47.512] L0RINDA: @bigdaas yep, that
1630[07:12:48.186] mariushav: @L0RINDA exactly
1631[07:12:50.148] frakty_zero: Leela's magic resource is 2/21
1632[07:12:56.051] kanchess: brilliant vid Aloril42 :D
1633[07:13:02.395] pierre_sf: !reverse
1634[07:13:02.608] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
1635[07:13:06.032] eval_bot Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
1636[07:13:13.323] alayant: Fail low now after the fail-high
1637[07:13:27.145] bigdaas: thanks @d0rus @L0RINDA
1638[07:13:27.183] d0rus: That reverse on moobot is anoying
1639[07:13:33.139] alayant: Crazy unstability
1640[07:13:35.425] jim_vs: !sux Stockfish
1641[07:13:35.691] Nightbot: Why Stockfish sux now?
1642[07:13:43.180] doppelbauer: ok, white back to reality :)
1643[07:13:51.230] NikolayKralev: sf IS DEAD
1644[07:13:51.634] enqwert81: Leela has Ng6
1645[07:13:53.223] gakuj505: 1-0
1646[07:13:56.778] theo77186Fr: Just crash already Kappa
1647[07:14:01.630] georgelvoinicel: @L0RINDA what is beyond my understanding is this veto system
1648[07:14:07.443] snipertooth: how many difference in the stockfishs tablebase
1649[07:14:10.315] ma_d____m_an: Sf moob ...ishe isnt forced to play Nc6
1650[07:14:24.308] Occyroexanthub: Leela likes Ng6 yes!!
1651[07:14:27.043] Occyroexanthub: homeLeela
1652[07:14:27.182] Aloril42: @kanchess Sounded like it would be less than 10 years before robots pass humans on table tennis by big margin.
1653[07:14:28.840] fresh_salsa_chips: moobbot Kappa
1654[07:14:55.125] d0rus: SF will think a long time now since it saw a big moob
1655[07:14:56.322] GrahamNZ: by Ferdy » Sat Sep 14, 2019 11:35 pm Enter engine name? Alliestein Rule 1: The code for training the NN, unique? (y/n) n Rule 2: The NN weights file, unique? (y/n) y Rule 3: The engine that executes the NN, unique? (y/n) n TCEC Season 16, NN uniqueness test summary: name: Alliestein Pass Rule 1: n Pass Rule 2: y Pass Rule 3: n Eligibility for participation: Fail A minimum of 2 rules should pass! Ref: Rules/sections 16.b, 16.c and 16.d https://tcec-chess.com/
1656[07:15:08.278] mrbdzz: SF moobs big time. Hopefully won't get a stroke tcecK
1657[07:15:12.480] gakuj505: 1.8 -->0.4 PogChamp
1658[07:15:12.893] josephviruses: Laual
1659[07:15:19.003] josephviruses: LUL
1660[07:15:29.857] GrahamNZ: http://www.talkchess.com/forum3/viewtopic.php?f=2&t=71822#p812105
1661[07:15:30.953] theo77186Fr: Still alive
1662[07:15:32.452] Archstro: Dreamfish woke up
1663[07:15:38.373] Occyroexanthub: LEela will find 24...Ng6 as it seems from my homeLeela divP net
1664[07:15:42.541] gakuj505: fish only fish Kappa
1665[07:15:43.730] nickpelling: It is a critical position, so a long think would be justified
1666[07:15:44.406] jim_vs: !ismoob 1.8 0.4
1667[07:15:45.674] Occyroexanthub: so SF will moob
1668[07:15:48.648] Occyroexanthub: let's see though
1669[07:15:49.570] josephviruses: RF now same as Leela's pv too
1670[07:16:02.277] theo77186Fr: SF already moobed LUL
1671[07:16:04.928] L0RINDA: @georgelvoinicel Don't think of it as a veto, think of it as the players being polled as to whether they are all okay with something being changed that shouldnt be. This is common in many amateur leagues in games and sports
1672[07:16:06.045] georgelvoinicel: SF is always confused by Lc0 moves in French.
1673[07:16:08.988] Occyroexanthub: :P
1674[07:16:13.605] Occyroexanthub: nice
1675[07:16:22.220] Occyroexanthub: how SF was blind all along it's strange
1676[07:16:22.332] fish_stock: well its an extraordinary decision and therefore requires unanimous agreement
1677[07:16:26.119] mrbdzz: RF at 0.00! PogChamp
1678[07:16:30.758] Occyroexanthub: but on the other hand redSF is 256 threads :P
1679[07:16:38.004] sirinial64: That was an epic moob by whitefish
1680[07:16:53.972] Martian_the_Martian: SF needs to take classes of LCzero on how to deal with such positions as in this game.
1681[07:17:04.801] HenriKar_Chess: Can we get a command for 'epic moob'?
1682[07:17:08.210] d0rus: RF saw it too so idk
1683[07:17:21.865] bigdaas: looks like it may just be an all-NN sufi?! wow
1684[07:17:23.899] Aloril42: Rarely see tcecM from tcecSF
1685[07:17:39.120] GrahamNZ: AllieStein is ineligible according to TCEC rules
1686[07:17:40.605] L0RINDA: @georgelvoinicel We can take a vote as to whether everyone in this chat should give me $500 for instance. Should that be a majority issue :P
1687[07:17:47.885] HenriKar_Chess: !hype tcecM
1688[07:17:48.076] Nightbot: aweSome
1689[07:18:00.827] mariushav: Allie has 3/4 wins against komodo. Crazy!
1690[07:18:13.230] georgelvoinicel: @L0RINDA hehe
1691[07:18:19.592] Occyroexanthub: K MCTS
1692[07:18:21.777] Occyroexanthub: not Komodo
1693[07:18:23.651] HenriKar_Chess: I guess Lc0 found a real opening novelty here
1694[07:18:30.782] josephviruses: Will leela become negative?
1695[07:18:34.529] L0RINDA: @georgelvoinicel Joking aside though, it's the same kind of thing, they were given the opportunity to make a gesture, basically
1696[07:18:38.646] sometimesithurts: sf is blind in fortresses and openings (the first 30 moves or so) but after move 30 it's evals are really stable
1697[07:18:52.446] L0RINDA: @georgelvoinicel For reasons we'll likely never know for sure, one person did not
1698[07:18:55.573] gakuj505: next year have redaliie
1699[07:19:13.057] bigdaas: which rule? @GrahamNZ
1700[07:19:24.585] kanchess: it's incredible Aloril42 had a great laugh of pleasure here viewing it together with some people
1701[07:19:25.047] alchemist888: clonellie
1702[07:19:29.664] GrahamNZ: did you read what I posted?
1703[07:19:39.553] HenriKar_Chess: Even _my_ homeleela (CPU LD2) finds Ng6
1704[07:20:01.419] Occyroexanthub: @GrahamNZ you have to post it to TCEC's discord maybe
1705[07:20:07.121] GrahamNZ: Enter engine name? Alliestein Rule 1: The code for training the NN, unique? (y/n) n Rule 2: The NN weights file, unique? (y/n) y Rule 3: The engine that executes the NN, unique? (y/n) n TCEC Season 16, NN uniqueness test summary: name: Alliestein Pass Rule 1: n Pass Rule 2: y Pass Rule 3: n Eligibility for participation: Fail A minimum of 2 rules should pass! Ref: Rules/sections 16.b, 16.c and 16.d https://tcec-chess.com/
1706[07:20:09.008] bigdaas: checking tcec rules @GrahamNZ
1707[07:20:15.447] bigdaas: ty
1708[07:20:45.721] Archstro: I think the vote is wrong because it puts the other competitors in an unfair situation. Voting no inherently looks bad, so you're kinda forced to vote yes.
1709[07:21:23.252] sirinial64: @Archstro hence the anonymity of the vote
1710[07:21:40.814] pierre_sf: !alliestein
1711[07:21:41.091] Nightbot: See !allie and !leelenstein
1712[07:21:45.868] Archstro: It's not anonymous, the TCEC team knows who voted what.
1713[07:22:00.074] jim_vs: only Anton knows
1714[07:22:07.532] jim_vs: I think
1715[07:22:12.797] pierre_sf: !leelenstein
1716[07:22:13.048] Nightbot: Leelenstein aka Dr. Leelenstein, aka The Doctor, is a 20x256 SE net built using games from CCRL, Sergio's 40 block variant, T30 + TCEC games. Novelties: Initalization + TP + GGT + SGDR + Gradient Noise created by jjosh https://www.patreon.com/jjosh see !ls
1717[07:22:14.616] ma_d____m_an: I know
1718[07:22:18.377] mikebridges: looks like leela's defended it
1719[07:22:29.256] digretion: just revisited from last season, still seeing engine doing wild eval swings and prone to crashes. wait, is that fish?!
1720[07:22:35.775] HenriKar_Chess: @GrahamNZ I guess the person that made that post disagrees on what the 'engine that executes the network' is
1721[07:22:38.586] Archstro: Even if nobody knows, you'd feel like an a-hole to vote no, even if it's in your right and I think it's the right decision.
1722[07:22:46.370] HenriKar_Chess: is that just the NN part? of the NN-part + the search?
1723[07:23:04.693] mrbdzz: Leela expecting exchange sac :O
1724[07:23:06.251] nickpelling: @mikebridges looks more like whitefish stopped dreaming
1725[07:23:21.690] Occyroexanthub: SF wants Rxg6 PogChamp
1726[07:23:27.649] jim_vs: personnaly I would not feel like an a-hole to vote no. And I am SF fan
1727[07:24:04.030] Martian_the_Martian: Would changing the dll be against the rules, that is the question.
1728[07:24:04.315] Archstro: I'd vote no even if it was about my own engine.
1729[07:24:05.563] sirinial64: The no vote is Stockfish dev Kappa
1730[07:24:17.863] radjehuty: Impressed by Allie, didn't think she'd do this well
1731[07:24:25.193] fresh_salsa_chips: @sirinial64 u sir are a genius SeemsGood
1732[07:25:17.426] nickpelling: Really don't know how this is going to play out. Hopefully not with a crash. :-(
1733[07:25:31.492] sirinial64: @fresh_salsa_chips I'll be happy to prove you wrong very soon Kappa
1734[07:25:35.548] Occyroexanthub: @GrahamNZ thanks for that CCC topic. i missed that. reading it and Crem's diagrams somewhat changes my opinion indeed http://www.talkchess.com/forum3/viewtopic.php?f=2&t=71822#p812105
1735[07:25:52.069] fish_stock: @Archstro SF abstained, lc0,komodo's,houdini,stoof voted yes scorpio and allie I'm not sure and wouldnt dream of speculating...
1736[07:26:14.752] positiev3xxx: Normally Leela would have +15 mins advantage about this time
1737[07:26:19.413] alchemist888: Change a library is not a replacement for the engine. Everything is fair here. What reason to say "no"? Knock out a competitor and have a chance at win? But why? Without strongest competition you are not the champion
1738[07:26:28.691] hiep_vu: Leela had a tough time
1739[07:27:14.972] radjehuty: Leela played way too many draws. I look forward to seeing her T60 net though
1740[07:27:27.212] Twipply: Inviting an "undisclosed expert" is amusing
1741[07:27:28.482] fish_stock: @alchemist888 someone voted no... but who?
1742[07:27:35.913] hiep_vu: @alchemist888 : are you sure the "champion" voted no ?
1743[07:27:40.221] GamBito93: @alchemist888: I'm with you. But they don't have the best rules here
1744[07:27:44.992] Twipply: As is trying to keep it a secret
1745[07:27:48.827] enqwert81: They have the same PV for the next 10 moves.
1746[07:27:49.351] Archstro: You can put the whole engine in a dll and change it whenever you want then.
1747[07:27:53.790] Twipply: especially when a bunch of people in chat already call them LCZ clones
1748[07:29:04.923] alchemist888: @hiep_vu I know Leela team voted "yes"
1749[07:29:16.431] Hanamuke: this looks like a position in which SF could have a tactical edge
1750[07:29:28.075] Twipply: @alchemist888 do you?
1751[07:29:37.766] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
1752[07:29:41.081] pierre_sf: A virus LUL @Archstro
1753[07:29:52.954] positiev3xxx: Making a change during the race/competition is always penalized in any type of sports. Just for an example, in F1, you have to start the race at the very end if you switch engine. So, whoever voted no has every right to do so. Congratz.
1754[07:29:57.809] xoto10: you can't put a whole engine into someone else's project
1755[07:30:08.123] alchemist888: @Archstro this is not the case
1756[07:30:53.297] alchemist888: @Twipply Yes, I think Allie is a clone of Leela, like Sugar is a clone of SF
1757[07:31:05.262] 1coghoti: @Archstro how out an en-source replacement
1758[07:31:09.271] Twipply: @alchemist888 I meant that you somehow know what LCZ voted
1759[07:31:10.489] Twipply: How so?
1760[07:31:24.588] Twipply: Yes, I'm aware there was a public vote that went with yes, but that does not mean it was followed
1761[07:31:46.847] Occyroexanthub: well same applies to all of you
1762[07:31:52.809] Occyroexanthub: that you said you will vote yes
1763[07:31:55.262] Twipply: Indeed it does
1764[07:31:57.111] Occyroexanthub: so we know nothing indeed
1765[07:31:58.748] Occyroexanthub: for sure
1766[07:32:06.902] Twipply: Although personally I gave Anton permission to reveal our vote
1767[07:32:12.754] Twipply: So at least that'd count as "known"
1768[07:32:20.312] Occyroexanthub: nope
1769[07:32:24.767] gakuj505: f*ck fish
1770[07:32:24.807] Occyroexanthub: we don't know you said that :P
1771[07:32:34.195] alchemist888: @Twipply conspiracy teory? :) Is not too much for some computer chess?
1772[07:32:37.347] sometimesithurts: you might knew anton won't reveal it :)
1773[07:32:37.517] Twipply: Anyone there at the time would know
1774[07:32:41.196] Twipply: I can say it again now if you like
1775[07:32:49.916] Twipply: Hey Anton, please reveal Stoof's vote about SF to the chat, thx
1776[07:33:07.843] L0RINDA: I was the 8th vote and voted no, because i like chat to be ansty
1777[07:33:07.976] pierre_sf: is alliestien completely open source?
1778[07:33:09.996] Occyroexanthub: to be honest i don't care about the voting so much but since it's just one "no" then ok i would like to know :P
1779[07:33:12.657] Twipply: Worth noting that he ignored me and didn't reveal it, but eh
1780[07:33:40.661] Occyroexanthub: Allie is completely open source
1781[07:33:57.832] alchemist888: but net is not
1782[07:34:00.885] sometimesithurts: what's your point now? @Twipply why is there 8 votes when there are 7 teams?
1783[07:34:07.780] LeonardoDiCaprino: the vote NO came from the one who was going to benefit the most from a YES. that way, that person can make sure that its not their fault and the system is rigged against them. rally chat. rally by their side! Kappa
1784[07:34:20.323] Aloril42: If tcecDD still not boring... excitement stays until game is over.
1785[07:34:33.953] Twipply: @sometimesithurts there's an extra vote 'cus Anton messed up the numbers accidentally, that's all
1786[07:35:06.835] gakuj505: HolidayCookie
1787[07:35:17.560] sometimesithurts: anton said it counted Komdo team's vote twice, as they had two engines.. and in the end of the day, it doesn't matter
1788[07:35:19.892] Twipply: Presumably if he'd messed up a no vote, then the entire result of the vote would've changed so he would've said
1789[07:35:25.079] Twipply: meaning it was an extra yes/abstain
1790[07:35:31.044] positiev3xxx: Thankfully, nowadays dead draws are fun to watch if SF is playing LUL
1791[07:35:34.693] Twipply: but it's also hard to create a vote for something that had no votes
1792[07:35:37.456] Twipply: miscounting is easier
1793[07:35:37.979] gakuj505: losefish PogChamp
1794[07:35:39.878] sirinial64: Stein @alchemist888 you can download the stein nets here on jjosh patreon for free, including the latest 10.2 net
1795[07:35:44.724] Twipply: so presumably it's actually 5/1/1
1796[07:36:02.527] sometimesithurts: it is I suppose
1797[07:36:11.754] sirinial64: !stein @alchemist888
1798[07:36:11.927] Nightbot: There is a description and links to papers at patreon.com/jjosh
1799[07:36:15.937] Martian_the_Martian: Reading the rules I'd say it's not obvious that crashes caused by a dll should count as an engine crash. Therefore I would give SF the benefit of the doubt, replace the dll and resume the crashed games.
1800[07:36:18.406] pierre_sf: where can I find a link to allies net?
1801[07:36:41.689] alchemist888: @sirinial64 are you sure that is latest Stein net>
1802[07:37:06.869] sirinial64: Jjosh told me so directly @alchemist888
1803[07:37:09.644] 228artyom228: #ilovelc0
1804[07:37:25.290] alchemist888: @sirinial64 ok then
1805[07:38:09.720] ジブ940 (jb940): @Martian_the_Martian the point was the dll was part of the sf build and part of the submission, but the ambiguity was given to vote for. I don't agree with letting authors decide, but they tried their best to handle it.
1806[07:39:17.594] digretion: remember the CompilerGate
1807[07:39:33.282] ジブ940 (jb940): I 'member
1808[07:39:35.668] chess_fan26: @ジブ940 one more crash and it is the end for sf
1809[07:40:00.318] Occyroexanthub: SF probably will not crash
1810[07:40:35.337] d0rus: Sf can always try again next season
1811[07:40:39.894] Occyroexanthub: i can't be sure but i think it will not
1812[07:40:43.684] talkshowwcharles: Hopefully not.
1813[07:40:55.918] saulzar: I think engine authors should be required to submit patents describing the workings of the system and proving uniqueness, written by lawyers so that no one can possibly understand
1814[07:40:58.897] fish_stock: it defo will
1815[07:40:59.796] Martian_the_Martian: Yes, I understand the dll was part of the submission, but it is part of the engine? So, therefore I say because the ambiguity, give SF the benefit of the doubt.
1816[07:41:05.679] chess_fan26: Yea I pray it doesn't
1817[07:41:06.753] ジブ940 (jb940): Yea. I'd like to see sf manage, but either it happens or it doesn't. For me either makes a pretty fair sufi. Does suck for people who desperately want to see sf though.
1818[07:41:22.126] LeonardoDiCaprino: you dont even need to know what happened in specific. this is TCEC chat. heres what to expect: people are mad that SF isnt 100/0/0
1819[07:41:22.957] ジブ940 (jb940): Fun* not fair, but fair works too.
1820[07:42:11.753] sometimesithurts: hasn't anyone tested the buggy version to see what is the crash rate?
1821[07:42:19.032] saulzar: crem clearly wants to DQ allie - that is interesting that Lc0 want to kill allie, defining the NN format as part of the NN is quite disingenous
1822[07:42:23.101] ジブ940 (jb940): @Martian_the_Martian but you enter a minefield that turns tcec into a testing ground. Oh a third party tool? Just update without testing, we can always revert and resume.
1823[07:42:50.279] mikebridges: sf's remaining task is not to crash now that it has failed to obtain a winning position
1824[07:42:55.249] saulzar: You would assume that the NN uniqueness would be defined by weights... at least I would think that is the obvious interpretation
1825[07:43:07.477] f1tof5: I just read through the discussion that Crem posted. He says Allie fails two of three criteria for uniqueness including unique NN engine. But ...
1826[07:43:19.689] nickpelling: @mikebridges the first edge-of-seat moment of many more to come
1827[07:43:22.856] Martian_the_Martian: They should make the rules more clear to avoid a minefield.
1828[07:43:23.628] ジブ940 (jb940): Was that crem who said that?
1829[07:43:25.782] saulzar: He redefines the NN uniqueness to suit his argument basically
1830[07:43:29.161] f1tof5: Surely the uniqueness of the engine is a combination of the NN plus search algorithm?
1831[07:43:38.283] ジブ940 (jb940): Anyway this argument happened months back
1832[07:43:44.608] Twipply: @saulzar What makes you think he wants Allie out?
1833[07:43:44.794] f1tof5: Sorry it wasnt crem.
1834[07:44:05.894] mikebridges: N+P+P is approximately = R
1835[07:44:06.346] Aloril42: I think it's good to share file formats instead of having many different formats.
1836[07:44:10.113] sometimesithurts: @saulzar so what if I train the net for one more epoch? is it unique?
1837[07:44:11.012] ジブ940 (jb940): Allie uses lc0 code to work with weights and work with cuddn, but just like the threading library, those are considered fundamentals
1838[07:44:11.281] alchemist888: Actually Allie faild 3 of 3 criteria
1839[07:44:16.180] f1tof5: It's the last person who posted in the discussion.
1840[07:44:26.129] nickpelling: 0.00 is exactly a draw
1841[07:44:34.174] oci82: @sometimesithurts what if I change the 10th decimal of one value in the NN file? :)
1842[07:44:34.970] saulzar: @Twipply He's made these fancy diagrams to redefine what NN uniqueness means...
1843[07:44:36.822] ジブ940 (jb940): But allie search and policy was custom coded, and the nets are custom too.
1844[07:44:37.572] f1tof5: But Allie's search algorithm is unique. To my mind it passes 2/3. That should be good enough.
1845[07:44:47.320] f1tof5: Agreed.
1846[07:44:58.915] fish_stock: Leela is a copy of AZ so thats not unique
1847[07:45:11.990] saulzar: @sometimesithurts Yeah well - I think all these rules are very fuzzy
1848[07:45:15.069] AntonMihailov: Hi everyone! Allie on a winning streak, strong showing
1849[07:45:19.952] f1tof5: Ok, I thought there was still some running controversy about that. Seems there isn't. Sorry for bringing it up.
1850[07:45:42.284] saulzar: @sometimesithurts Just like if I rename all the variables in my code, is my engine unique?
1851[07:45:43.270] alchemist888: @ジブ940 (jb940) net is mostly T30 because trained on T30 games with the same training process
1852[07:45:59.327] Twipply: @saulzar pretty sure since the topic title is about CHANGING the rules, he's not trying to get Allie out
1853[07:46:05.358] digretion: the efforts to train an unique NN model is tremendous. from microcomputing perspective let's just share public models
1854[07:46:07.676] xoto10: @ジブ940 "turns tcec into a testing ground" That's a theoretical risk, but I don't believe it's a big risk in practice. irl there are engines that regularly crash and engines that get bugs fixed quickly - I think that's more to do with size & strengths & attitude of team than tcec rules
1855[07:46:10.136] Twipply: Why argue that Allie violates the rules, and then say you want the rules changed
1856[07:46:12.820] oci82: yes, simply allow all engines to play, competition will do the rest.. my opinion.. since there is nothing to be won
1857[07:46:14.228] ジブ940 (jb940): @saulzar there has to be some fuzzyness. Ab engines are 80% similar too. The 20% makes them play very different though.
1858[07:46:14.875] Twipply: if your goal is for Allie to get dq'd
1859[07:46:19.856] m7md_wfa: What is going with T60 btw ?
1860[07:46:33.876] sometimesithurts: uniqueness is vague .. even the search algorithm uniqueness .. what if change a single line of leela's search? and train for another epoch.. then I have a unique search and a unique net .. bingo here is my new unique enging
1861[07:46:57.591] Occyroexanthub: what is wrong with T60?
1862[07:47:03.499] saulzar: @Twipply Given that it seems it occurred a while back, whoever brought up this topic after all this time seems to have an agenda perhaps?
1863[07:47:14.061] Archstro: T60's progress seems a bit too slow
1864[07:47:22.921] Twipply: @saulzar that agenda being about TCEC admins not doing stuff
1865[07:47:23.304] Occyroexanthub: nah
1866[07:47:34.753] ジブ940 (jb940): It's bigger and in a lower LR, yea it's gonna be slower
1867[07:47:37.245] digretion: the fabled NN-wall
1868[07:47:56.210] AntonMihailov: @Twipply is this discussion about the Talkchess thread?
1869[07:48:00.788] LeonardoDiCaprino: lol one day the admins are gonna shut this whole thing down to avoid the headache of dealing with the angry viewers
1870[07:48:02.721] saulzar: @ジブ940 In the end it always will come down to some judgement - of course everyone likes a good rules dispute
1871[07:48:05.816] Twipply: @AntonMihailov Indeed
1872[07:48:22.511] Twipply: Dunno who brought it up though
1873[07:48:23.110] m7md_wfa: but that slow ?!
1874[07:48:39.236] ジブ940 (jb940): @LeonardoDiCaprino drama drives viewers. Nah it's all good OpieOP
1875[07:48:40.126] AntonMihailov: @Twipply I can talk about the topics touched there for hours... I cannot do it on talkchess though as my account was never activated
1876[07:48:52.904] m7md_wfa: it is stuck in 2400 for like 2-3 weeks
1877[07:48:54.350] Occyroexanthub: TCEC needs much better rules for NNs that's for sure :P
1878[07:48:57.275] LeonardoDiCaprino: @ジブ940 aaaaaaah!
1879[07:49:12.307] Twipply: @AntonMihailov I have nothing to do with that thread my dude, doesn't matter to me if you reply there or not
1880[07:49:14.357] Twipply: It's crem's thread
1881[07:49:20.662] Occyroexanthub: to clarify with more strict way when a NN is unique or not.
1882[07:49:22.833] raff1st: it would be much more interesting final if human programmed SF played machine learning NN, like it's more interesting to see Bayern vs Real than Real vs Barca Kappa
1883[07:49:30.155] Twipply: Although I could go tell him that you can't reply there
1884[07:49:47.788] ジブ940 (jb940): In the end, the stein networks have been deemed sufficiently different (and the engine makes different opening moves already, so atleast for sufi its not gonna be same same), and the search/policy part of the engine allie is entirely new and handwritten
1885[07:49:56.877] ジブ940 (jb940): So to me, that's just 2 out of 3 pass.
1886[07:50:07.387] AntonMihailov: @Twipply we tried to have better rules at the start of the season regarding uniqueness. That never finished in a complete rule set, thus was not pushed further. The rules remained the same, with some vagueness and lot's of responsibility for me to decide what is and what is not unique
1887[07:50:32.223] Twipply: Noted
1888[07:50:59.053] m7md_wfa: !rules
1889[07:50:59.199] Nightbot: To see the rules, check the rules button in the GUI top left or see https://wiki.chessdom.org/TCEC_Season_16_Rules nadedKing
1890[07:51:04.097] fish_stock: what if supporters of the lc0 project withdraw their support because their generosity is being abused. by chancers and wide boys
1891[07:51:07.369] saulzar: I'm not really sure you can make rules which have no human interpretation without getting very very silly
1892[07:51:07.933] ジブ940 (jb940): @AntonMihailov but hasn't that always been the case? Even AB engines share like half of their code atleast. And like 85% of their functionality. Yet the other 15% makes a big difference.
1893[07:51:26.366] ジブ940 (jb940): @AntonMihailov I'm arguing that that vagueness is fair, to add.
1894[07:51:29.584] josephviruses: @fish_stock that's a concern
1895[07:51:45.583] AntonMihailov: In general the whole uniqueness question was initiated by Crem regarding DeusX. That was the spark of the discussion
1896[07:51:58.980] chess_fan26: Imagine if money was on the line!
1897[07:52:02.997] Twipply: @ジブ940 personally I can't understand why we care what the training code is
1898[07:52:12.355] Twipply: Everyone uses cutechess, that ain't unique
1899[07:52:13.296] saulzar: For example that diagram that crem made - in 2 years time some other architecture of engine will show up which doesn't match any part of that diagram
1900[07:52:25.975] AntonMihailov: That was when I pointed that my technical knowledge is not to the standard to analyze in detail uniqueness and that there should be guidelines
1901[07:52:27.450] xoto10: @saulzar Agreed. If it's easy to make more specific rules, ok, but I think it's important to make the spirit clear and leave some of it up to the TD's discretion
1902[07:52:27.812] Choca_13: The problem with NN based engines is that you need a lot of time and/or processing power to get a good NN
1903[07:52:30.856] chess_fan26: !DeusX
1904[07:52:30.971] m7md_wfa: @chess_fan26 maybe there is
1905[07:52:31.085] Nightbot: DeusX is a neural net trained by Albert Silver. Unlike Leela Chess Zero, which determines the weights through iterated self-play, DeusX was trained through supervised learning on a database of human games with around 3 million games. DeusX uses the Lc0 engine.
1906[07:52:46.360] lado933: can leela blunder here?
1907[07:52:51.237] ジブ940 (jb940): @Twipply because i guess it alters fundamentally how the nets turn out. But yea.
1908[07:53:16.098] AntonMihailov: For me common sense is never common, nor some people have sense... All working systems are based on at least a basic rules set to be followed
1909[07:53:41.549] xoto10: Is it possible to analyse how much difference there is between the moves various engines choose?
1910[07:53:41.907] Twipply: @AntonMihailov so who was that 'undisclosed expert' crem mentions?
1911[07:54:10.608] AntonMihailov: Crem indeed deleted the discussion, that does not mean it was not saved and continued. It was not finished for sure
1912[07:54:19.972] Twipply: Run engines A and B on N positions and see how many match?
1913[07:54:37.248] Twipply: Not hard to measure, but perhaps hard to say if the answer is relevant
1914[07:54:41.551] ジブ940 (jb940): @Twipply I guess since training isn't (or doesn't have to be) part of the submission, it's just separate, you can just argue (just like with AB), that it's a very important part someone coded themselves. Some ab engines also kinda play same same, the importance is its originally made.
1915[07:54:52.903] Twipply: @AntonMihailov but who was the 'expert' ?
1916[07:54:56.007] AntonMihailov: @Twipply actually not one "expert", I try to get the opinion of as many professionals as possible,
1917[07:55:20.505] Twipply: crem refers to one that joined your discussion particularly
1918[07:55:23.352] Twipply: Just curious who that one is
1919[07:55:39.128] f1tof5: @Twipply Are engines deterministic enough for that test to work?
1920[07:55:45.446] AntonMihailov: @Twipply I wish more have joined in time before it disappeared
1921[07:55:53.953] Twipply: @f1tof5 sure why not who knows
1922[07:55:59.931] josephviruses: SF will win against Scorpio in next game and it is game over for leela if she doesn't keep on winning.
1923[07:56:03.406] Occyroexanthub: with 1 thread and fixed nodes they might be deterministic enough
1924[07:56:15.502] xoto10: @Twipply Yes. well, if we had the data, presumably some engines would be closer together than others. That way we're measuring hard results, not making vague judgements about the build process and the built sw/data
1925[07:56:23.178] saulzar: Any kind of concrete rules for uniqueness will be very easy to sidestep once you know what they are..
1926[07:56:35.920] Twipply: Seems odd that 'experts' would be able to join if the topic was meant to be a secret as Crem claims
1927[07:56:59.473] AntonMihailov: To sum up: I see why Crem is not happy, but the work is not done. The question remains open and until a solid solution is found I will continue to carry the responsibility of uniqueness
1928[07:57:09.398] ジブ940 (jb940): @xoto10 but that's also unfair. Two engines could play similarly, but made entirely by different people, functioning entirely different, hell, RF and Leela could be too similar.
1929[07:57:13.673] xoto10: @f1tof5 just run the test more times
1930[07:57:16.217] noobs_killer: Leela is too strong for SF, SF needs 1000 cores!!
1931[07:57:16.313] m7md_wfa: Do you think Aille is stronger then Leela ?
1932[07:57:16.949] AntonMihailov: And that is a huge responsibility, given the history of computer chess
1933[07:57:44.930] Twipply: @AntonMihailov you know, I do think that rules committee channel should be opened up for view by at least engine authors
1934[07:57:55.040] noobs_killer: @m7md_wfa , Allie> Lc0 >SF!
1935[07:58:08.401] Lc0fan: the uniqueness issue has been a theme hroughout TCEC history.
1936[07:58:14.110] Twipply: Like you said, you're personally not an expert at the NN stuff
1937[07:58:22.287] Twipply: I'm not personally aware of any tcec admins that're huge experts at this
1938[07:58:23.807] ジブ940 (jb940): As long as @Twipply doesn't get viewing access as he wants to expose every anonymous person Kappa
1939[07:58:27.157] xoto10: @ジブ940 sure, I was thinking of those 2. But that would give us a guide as to how much difference is ok. Presumably other tests would give a guide as to how close is not ok, perhaps Sugar vs sf
1940[07:58:27.313] lado933: @noobs_killer keep dreaming
1941[07:58:36.252] CatoTheYounger_TCEC: fwiw I think we're just overthinking this
1942[07:58:36.842] AntonMihailov: @Twipply I think it should be , some privacy issues though. I will think of a way
1943[07:58:37.213] Twipply: @ジブ940 ShHhHhHhH
1944[07:58:40.487] saulzar: Don't require an engine author to make sensible judgements about these things - it's not unique to chess
1945[07:58:55.468] TCEC_Spectator: Again, I have to side with CatoTheYounger_TCEC
1946[07:59:00.276] noobs_killer: @lado933 , this is evidence!
1947[07:59:14.049] m7md_wfa: I think it is not over yet we should wait until the end of the divp
1948[07:59:15.320] AntonMihailov: @CatoTheYounger_TCEC I agree here, it is overthinking, just I was personally mentioned and felt it is my duty to respond
1949[07:59:16.147] chess_fan26: @m7md_wfa it isn't!
1950[07:59:28.921] Muzgrob: I'm here for computer chess. Not drama. So whatever the admins decide is fine by me.
1951[07:59:32.408] xoto10: @CatoTheYounger_TCEC Agreed. It's TCEC competition, you guys make the rules and decisions, end of.
1952[07:59:40.107] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
1953[07:59:44.021] ジブ940 (jb940): @xoto10 but I think it's just fundamentally wrong. It's the code that matters, not the play. Engines should just be built by different methods with different coding perspectives
1954[07:59:49.630] Twipply: @Muzgrob admins decided that you can't watch anymore, sorry
1955[08:00:00.729] f1tof5: Here's a possibly fun way of resolving the issue of identical NNs with different weights. TCEC could have a bonus competition between them, and the winner enters that season.
1956[08:00:05.734] RubiChess: "Come for the chess, stay for the Drama"
1957[08:00:08.213] Muzgrob: But I'm not banned yet?
1958[08:00:08.476] Schumacherx: Not an exciting season so far tbh. Are the engines getting too strong or it's the opening choice. Perhaps use more aggressive lines. I can always help. LUL
1959[08:00:09.265] ma_d____m_an: It is better for SF to find a perpetual to stop leela from extending the game 3 more hours maximizing crash chances
1960[08:00:13.517] halberder: I would like to see a rule set that allows a hybrid stockfish/leela engine that could be stronger than either
1961[08:00:20.190] ジブ940 (jb940): @AntonMihailov overthinking is twitch chats' responsibility. Come on Anton LUL
1962[08:00:22.415] xoto10: @ジブ940 fair enough
1963[08:00:24.019] AntonMihailov: @CatoTheYounger_TCEC we do our best effort, sometimes there might be failures along the way. But with the team that has formed as core one in the past years I am confident that all major pitfalls can be avoided and right solutions are always on the table
1964[08:00:37.895] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
1965[08:00:44.631] saulzar: If TCEC chat can't make a controversy from nothing who can?
1966[08:00:47.336] Twipply: I propose that everyone listens to me from now on
1967[08:00:57.767] Twipply: I have some great ideas to turn this into an Ataxx tournament instead of chess
1968[08:00:58.100] kkeronen62: We admins just try to run TCEC best we can.
1969[08:01:00.239] Twipply: I love that game
1970[08:01:03.631] chess_fan26: @AntonMihailov please spare sf for the crashes
1971[08:01:08.266] ジブ940 (jb940): @saulzar I feel personally offended of you lowballing my capacities to create drama.
1972[08:01:13.844] CatoTheYounger_TCEC: we exist to promote computer chess...we want to see chess of the highest quality...we don't like clones, we do like originality... so how can you measure originality objectively? with move-similarity testing, what else
1973[08:01:15.632] saulzar: @Twipply Seconded
1974[08:01:16.311] digretion: when you're champ, people listen
1975[08:01:28.551] Twipply: @saulzar Knew I could count on your vote,thanks baby
1976[08:01:30.652] AntonMihailov: @chess_fan26 my wish is never to see any engine crash....
1977[08:01:37.268] chess_fan26: !admins
1978[08:01:37.600] Nightbot: TCEC admins in S16: Aloril42 , jjoshua2 , kkeronen62 nadedKing
1979[08:01:48.481] ジブ940 (jb940): @AntonMihailov have you tried going blind? Kappa
1980[08:02:03.303] Jeydra: !votes @AntonMihailov did K team vote twice or is there a miscount?
1981[08:02:10.744] lado933: @CatoTheYounger_TCEC some clones give u higher quality than originals
1982[08:02:35.267] digretion: the Chinese would second that
1983[08:02:38.145] AntonMihailov: @Jeydra I unintentionally put the number 8 instead of 7, but that helped conceal the true vote forever
1984[08:02:39.236] atanh: Come for the visually crushing chess, stay for the viscerally crashing champions.
1985[08:02:48.070] CatoTheYounger_TCEC: if they are better than the originals that is measured through Elo
1986[08:03:41.822] Twipply: @AntonMihailov I propose that engine authors are told what the votes were, otherwise there's no checking process behind the result
1987[08:03:45.434] Twipply: You could make up anything you want
1988[08:03:54.238] Jeydra: @AntonMihailov what's the actual vote then? Since I'd rather the !s16wiki article be correct
1989[08:03:54.686] gakuj505: ResidentSleeper
1990[08:03:57.478] ジブ940 (jb940): @AntonMihailov but hey, I know chat or people can be overwhelming sometimes, we just all have opinions and want to do our say and be heard. But the viewer counts and the investment of people in the tourney do show: the team is doing an amazing job. I hope the team never takes the eagerness of chat personally. There's proof you're doing great.
1991[08:04:05.605] AntonMihailov: I hope I managed to transfer my thoughts here @Twipply regarding the talkchess discussion
1992[08:04:23.970] Twipply: @AntonMihailov better to transfer them to crem honestly, it's hit thread and all that
1993[08:04:29.116] Twipply: I don't often post on talkchess
1994[08:04:41.324] Twipply: his* thread
1995[08:04:45.934] AntonMihailov: @Twipply of course I will drop him a note
1996[08:04:55.279] noobpwnftw: I think not Kappa
1997[08:05:06.929] noobpwnftw: it's already titled "failed" so the rest is just whining
1998[08:05:15.045] Twipply: Oh hey noobpwnftw had issues in that thread, too
1999[08:05:15.353] xoto10: @jeydra the vote result was a veto, just let it go
2000[08:05:24.880] aartdappels: @AntonMihailov my personal problem with the allie submission is that the net is starting to use T40 data, and now planning to use even more (10M) T40 games. to me that means that allie is moving towards leela, and not away. what are your thoughts?
2001[08:05:32.175] chess_fan26: @AntonMihailov I think before the division starts it should be made sure that all engines are stable
2002[08:05:36.144] Robbieh0102: why is depth 1
2003[08:05:39.407] ジブ940 (jb940): @noobpwnftw you're titled "noob" but somehow we value your opinion too Kappa
2004[08:05:54.861] noobpwnftw: my issue is simple, next time when some shit happens, noob him
2005[08:05:59.018] chessinprogress: Leela is holding the French reverse PogChamp
2006[08:06:01.193] Twipply: I have no opinions, only facts
2007[08:06:03.215] chess_fan26: !anton
2008[08:06:03.440] Nightbot: Anton Mihailov is our intrepid leader!
2009[08:06:05.689] ジブ940 (jb940): Noob him 4Head
2010[08:06:23.039] f1tof5: @chess_fan26 They do testing. But this SF crash happens once every dozen or so games.
2011[08:06:36.293] kingscrusher: Some stuff about neural networks in this video: https://www.youtube.com/watch?v=Fn589zeMij4
2012[08:06:56.135] chess_fan26: @f1tof5 then how come it happened twice here
2013[08:07:25.422] f1tof5: Twice out of 22 games. So 1/11 so far. It didn't happen during testing.
2014[08:07:45.232] f1tof5: It's an intermittent bug. Simply wasn't picked up.
2015[08:07:58.515] noobpwnftw: so it happens with many of things I do, I'm open to discussions but eventually I decide whether to make adjustments or not, it's very simple like it or not problem.
2016[08:07:59.595] Twipply: @kingscrusher "There were some crazies that thought great big neural networks ... could learn everything"
2017[08:08:08.534] Twipply: It's almost like we don't all have great big neural networks inside our heads that already prove this
2018[08:08:15.503] Aloril42: Testing was 10 games, duration about 3-4 DivP games worth.
2019[08:08:18.142] Twipply: how crazy of an idea indeed
2020[08:08:29.544] kingscrusher: i think our neuralnetworks are multi dimenional though
2021[08:08:50.124] noobpwnftw: then I tell him I'm allowed to be grumpy so take that Kappa
2022[08:08:51.206] Twipply: You might say LCZ's DCNN is multidimensional, too
2023[08:08:53.575] lrmead: @kingscrusher Eh? There are but 3 spatial dimensions ...
2024[08:08:58.040] Jeydra: @xoto10 I wonder what you thought my motivations are. I'm writing the goddamn Wikipedia article, of course I'm going to go for accuracy right?
2025[08:09:20.619] mariushav: Sure, the Pitts-McCulloch (ish) perceptions were proposed early on, and later the community learned that with several layers, one could classify non-linearly separable data too
2026[08:09:23.990] kingscrusher: i mean i think the modelling of neural networks if very simplistic compared to what we use
2027[08:09:37.599] Twipply: Probably
2028[08:09:38.968] AntonMihailov: @noobpwnftw hi, nice to see you, Just sent you an email
2029[08:09:44.981] lrmead: @kingscrusher Ok, that may be ...
2030[08:09:52.318] mariushav: But gradient methods were not computationally efficient for quite a while longer
2031[08:09:59.324] chess_fan26: @kingscrusher make a video on the games in which Allie destroys K and H
2032[08:10:03.770] Twipply: But rand()%6 is also a simplistic model of a die roll, yet it contains every single aspect that you care about
2033[08:10:12.446] Jeydra: There's a line in the article right now that says the vote was 6 yes 1 abstain 1 no, and if only 7 teams voted, this line must be wrong
2034[08:10:20.833] kingscrusher: another nn video : https://www.youtube.com/watch?v=UTfQwTuri8Y
2035[08:11:05.323] m7md_wfa: why it should be ?
2036[08:11:34.297] Occyroexanthub: @Jeydra look up he already answered you
2037[08:11:52.850] kingscrusher: blimey .. turing mentioned reinforcing learning
2038[08:11:55.423] kingscrusher: apparently
2039[08:12:05.047] lrmead: @kingscrusher Do you think that Allie has leapfrogged over Leela and SF or can it lose in a head to had matchup?
2040[08:12:17.869] sometimesithurts: !cdb
2041[08:12:19.291] chessdbcn (42w) • No moves, position queued for analysis. <ChessDBCN>
2042[08:12:27.604] Occyroexanthub: 23 games=> sssssssssssssssssssssss Kappa
2043[08:12:30.697] AntonMihailov: I might have missed some of the discussion above, as my internet here is weak. Tonight I will have better connection, as usual I will be happy to answer any suggestions, questions, etc
2044[08:12:32.340] kingscrusher: im torn about covering alliestein games.. it seems like... leveraging training from leela but witout credit to leela
2045[08:12:34.332] Jeydra: @Occyroexanthub I don't see it ?
2046[08:12:44.624] kingscrusher: should i be covering alliestein on youtube?
2047[08:12:52.715] f1tof5: Yes you should.
2048[08:12:53.800] chess_fan26: Yea
2049[08:12:56.403] tinduz_r: Yes.
2050[08:12:56.682] saulzar: @aartdappels Open source sucks huh. Make everyone work from scratch
2051[08:13:02.326] kingscrusher: but it seems to leverage the leela community effort
2052[08:13:05.052] f1tof5: And the author certainly credits Leela.
2053[08:13:06.824] Jeydra: Cover interesting games, regardless of who played them
2054[08:13:08.272] lrmead: @kingscrusher Yes
2055[08:13:26.863] tinduz_r: Then put that opinion at the start of your video but cover the chess.
2056[08:13:34.644] sirinial64: Welcome to TCEC - Top Chess Engine Controversies freshly delivered every day ! Kappa
2057[08:13:40.527] kingscrusher: well if they credit leela then maybe
2058[08:13:40.639] Occyroexanthub: he said that he wrote this by mistake but it's better this way to keep results a secret
2059[08:13:44.807] kingscrusher: i mean if they keep that credit
2060[08:13:50.634] Aloril42: Credit should be given. Sharing resources and ideas should be good for engine strengths.
2061[08:13:54.193] f1tof5: But it is sufficiently different from Leela. AB engines in tcec leverage from SF and others, which is fine. Same goes for neural nets.
2062[08:13:59.710] saulzar: Allie is a great project, I feel like as soon as any of these projects seem like they might start winning then the daggers come out..
2063[08:14:01.164] Jeydra: Yeah, but then what do I put in the article now?
2064[08:14:05.563] aartdappels: @saulzar huh? nah, I just don't want to see engines with the same ideas of chess because they were trained on the same data?
2065[08:14:05.671] kingscrusher: i didnt realise even Turing mentioned self learning
2066[08:14:12.309] Twipply: Every AB engine there is takes advantage of SF effort, just saying that alone doesn't mean much
2067[08:14:18.881] kingscrusher: according to this video, he reckons turing mentioned it: https://www.youtube.com/watch?v=UTfQwTuri8Y
2068[08:14:29.423] Occyroexanthub: @saulzar thne include Sugar, Cfish etc as they are open source huh?
2069[08:14:36.342] Occyroexanthub: this logic is dead wrong for TCEC
2070[08:14:40.323] Twipply: Self learning was around a long time before AZ I'm sure
2071[08:14:41.329] Occyroexanthub: *then
2072[08:14:46.922] Twipply: even in the context of playing games
2073[08:14:59.363] Occyroexanthub: for Backgammon from early 90s
2074[08:15:07.561] not_a_zombie__: i havent been here for much of div p and wtf is going on here
2075[08:15:07.791] Twipply: TD-Gammon
2076[08:15:08.099] Occyroexanthub: without it bot would suck
2077[08:15:08.790] saulzar: @aartdappels Fair enough - it is clear that LS (and also Allie) have a great dependency on Lc0 ... (and always did...)
2078[08:15:11.633] Occyroexanthub: *bots
2079[08:15:14.812] tinduz_r: Personally, Allie's "concept" is different enough from Leela though it uses the same data. That makes it unique enough. Leela license dictates just putting in the license in the builds which Allie does.
2080[08:15:31.431] Jeydra: Ah damnit I'll just put there was a vote and someone vetoed it
2081[08:15:35.889] f1tof5: KC, AllieStein is a unique development. I really think you should run its games. Some of them are very instructive.
2082[08:15:37.984] not_a_zombie__: allie in the lead, stockfish crashing?
2083[08:15:40.309] josephviruses: @not_a_zombie__ you missed a lot of drama Kappa
2084[08:15:40.319] chess_fan26: How much is Allie better than leela?
2085[08:15:42.955] kingscrusher: its an engine eat engine world out here
2086[08:16:03.599] saulzar: @Occyroexanthub Sure, it's a matter of degrees..
2087[08:16:14.381] f1tof5: @chess_fan26 We have no idea if it is. Hence this tournament.
2088[08:16:19.505] lrmead: @chess_fan26 That would need a head to head matchup and we might get one in the SuFi
2089[08:16:47.743] kingscrusher: did u guys know there might be a technological undertone to hitchikers guide to galax and 'dont panic' maybe influenced by 'dont crash' philosophy of erlang
2090[08:16:49.362] tinduz_r: It's like you have so many open data out there who others painstakingly made, but different people can come up with different results from it.
2091[08:16:50.360] Twipply: @kingscrusher I suggest more videos of you swooning over Stoofvlees and outside passed pawns with unopposed bishops
2092[08:16:55.624] saulzar: @Occyroexanthub There's no fair metric for defining this stuff, they're all derivatives - but how much.
2093[08:16:56.719] Twipply: I may be bias though
2094[08:16:59.861] noobpwnftw: @AntonMihailov you fall for my trolling apparently, email responded Kappa
2095[08:17:01.784] kingscrusher: the guy was personal friends with Allan Kay
2096[08:17:03.100] mariushav: If an engine gets relegated, will all the games it played be discounted?
2097[08:17:03.247] trajkoskigoran02: When will be released new version of Alliestein?? I mean this version wich play this tournament??
2098[08:17:16.221] chess_fan26: @f1tof5 @lrmead here both are h2h equal
2099[08:17:16.629] kingscrusher: Dont panic = Dont crash
2100[08:17:21.132] Occyroexanthub: it's already "there"
2101[08:17:27.506] noobpwnftw: it will crash, don't worry
2102[08:17:28.999] Occyroexanthub: i.e you can have it
2103[08:17:33.833] mariushav: From crashes I mean
2104[08:17:39.938] lrmead: @chess_fan26 Yes, but too few games; sufi is 100
2105[08:17:47.867] kingscrusher: Allan Kay is a friend of Joe Armstrong
2106[08:17:57.774] kingscrusher: Allan Kay also a friend of Douglas Adams
2107[08:18:02.691] gakuj505: PogChamp
2108[08:18:07.638] Dabbler1: is this the reverse of the game leela won with white?
2109[08:18:09.712] josephviruses: @Twipply gotcha! You watch @kingscrusher 's videos too apparently Kappa
2110[08:18:22.031] lrmead: !reverse
2111[08:18:22.236] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
2112[08:18:25.591] eval_bot [12:18:55] Game #64 "LCZero v0.22.0-nT40B.4-160" vs. "Stockfish 190826": [book] 9. Nf3 b5 10. h4 b4 11. h5 h6 12. Rh3 bxc3 13. bxc3 Nab8 14. Kf1 Nc6 15. Kg1 Rc8 16. Be3 Qa5 17. a4 Qd8 18. Rg3 a6 19. Nf4 cxd4 20. cxd4 Nb6 21. Bd2 Nc4 22. Bc3 Ne7 23. Nh4 Rb8 ... 1-0
2113[08:18:38.926] josephviruses: @Dabbler1 yes. But SF diverged immediately
2114[08:18:40.199] fish_stock: in a sufi leela would destroy allie
2115[08:18:42.518] kingscrusher: the Babel fish reminds me of Contract Checking messaging in Joe Armstrong Erlang.. language indpendence
2116[08:18:42.890] xoto10: @Jeydra To be more constructive, it seems clear to me the vote was 5/1/1.
2117[08:18:59.211] lrmead: @fish_stock That remains to be seen ...
2118[08:19:04.376] chess_fan26: BTW there is a version of Allie waiting which is better than the current version playing here!
2119[08:19:06.464] Dabbler1: thanks joe
2120[08:19:22.682] lrmead: @chess_fan26 How do you know it is better?
2121[08:19:24.301] josephviruses: Yw
2122[08:19:27.624] fish_stock: No it cannot possibly win
2123[08:19:45.362] chess_fan26: @lrmead cuz it is latest
2124[08:20:03.054] lrmead: @chess_fan26 Means nothing to me ...
2125[08:20:03.096] Occyroexanthub: till now this divP Leela is the only undefeated and has defeated SF once, with SF's being the only loss. Last year this was again repeated with LEela being whole divP undefeated and SF had only 1 loss from Leela.
2126[08:20:17.374] Occyroexanthub: not a coincidence i guess. Leela too good vs SF at H2H
2127[08:20:17.954] not_a_zombie__: allie's not unique in "improving over time"
2128[08:20:23.800] Occyroexanthub: but not against other engines this year
2129[08:20:34.238] Occyroexanthub: last year LEela was +4 from 3rd Komodo
2130[08:20:44.027] Occyroexanthub: this year Allie got some good boost
2131[08:20:56.026] digretion: H6.03 being the "standard candle"
2132[08:21:08.644] enqwert81: Leela is clearly the best engine here but too solid well to score in a round robin.
2133[08:21:18.578] lrmead: Allie seems to be leaving Leela in the dust here ...
2134[08:21:25.971] Occyroexanthub: Leela too cautious ? :P
2135[08:21:34.443] Archstro: Leela learned chess by playing against the best engine of all time so she thinks everyone out there is a tough opponent. She's not taking any risks.
2136[08:21:46.488] m7md_wfa: This divp is really harder than the previous one
2137[08:21:52.056] nickpelling: @lrmead Allie has had a bit of luck, I think it's fair to say
2138[08:21:53.509] mariushav: Not sure this is the right place to ask, but in the training of zero type engines, how much of a stochastic element is added in the gradient methods? I would think you would need quite a bit of exploring lines with estimated low expectations
2139[08:21:57.254] tinduz_r: It took the "engine that never loses" a bit far and misses some wins in the process now.
2140[08:21:59.436] Occyroexanthub: she obviously needs contempt for TCEC
2141[08:22:31.271] josephviruses: 8 don't know but it seems this net likes simplifying early and relying on its opponent blundering in endgame
2142[08:22:34.762] josephviruses: I
2143[08:22:37.068] lrmead: Leela main flaw is still missing draw lines such as perps or stalemates
2144[08:22:58.784] chess_fan26: @Occyroexanthub Leela should play on 100 contempt
2145[08:23:02.014] L0RINDA: @mariushav It's absolutely the right place to ask! I'm not in a good position to answer, other than I remember initially the temperature was a bit high in training which led to the infinitely long endgames :P
2146[08:23:10.102] m7md_wfa: All the engines here are strong but there are some inaccuracies from the last 2
2147[08:23:15.327] josephviruses: Yeah. She failed to convert that KMC game because of that @lrmead
2148[08:23:33.510] atanh: !faketemp
2149[08:23:34.619] Nightbot: The current GPU temperature is 87 Kelvin KappaWealth
2150[08:23:50.525] lrmead: And how does a computer program have a "temperature"
2151[08:23:58.290] m7md_wfa: Aille was going to fail as well but she was lucky there
2152[08:23:59.052] lrmead: @josephviruses I saw that ...
2153[08:24:15.507] chess_fan26: @lrmead jjosh personally told me 2
2154[08:24:37.290] Dabbler1: i love leela for throwing her pawns to the wind for the good of the attack, but I understand she has to evolve and be more solid :D
2155[08:24:38.367] L0RINDA: @lrmead As I understand it, it's the amount of randomness it throws in
2156[08:24:50.322] josephviruses: I guess Leela's strategy here is to drain SF's time and let it lose by hanging Kappa
2157[08:24:53.031] romanob: Leela has officially killed SF ??
2158[08:25:11.767] mariushav: @L0RINDA I can imagine it must be a fine line between exploring too scarsely and not getting convergence.. I guess the question of tuning such hyperparameters, selecting a good way to discretize the state space etc is not something we have a perfect solution to
2159[08:25:13.369] halberder: if sf was smart it would offer draw before it crashes
2160[08:25:15.994] lrmead: @romanob It beat SF fair and square
2161[08:25:20.656] L0RINDA: @lrmead The higher the temperature, the more it drifts from what it thinks is the right move in training
2162[08:25:36.669] Dabbler1: @halberder lol
2163[08:25:46.054] L0RINDA: @mariushav yeah, i think the joke goes that we'll need a NN to tune the parameters in the end. Not quite sure how much of a joke that is though in the long run ;)
2164[08:25:47.235] Aloril42: This game is not yet over, though it's starting to look like tcecDD
2165[08:25:48.962] ジブ940 (jb940): NNs with self play learn by trial and error, so to say. She learns something, and then finds better moves. But sometimes to find even better moves, she has to try them first to see their effects. Temperature is this randomness
2166[08:26:09.808] josephviruses: @romanob leela won the reverse of this game
2167[08:26:18.487] lrmead: @L0RINDA But how is that? I would constitute human interference ...
2168[08:26:31.944] xoto10: Could leela blunder if low on time?
2169[08:26:42.187] L0RINDA: @lrmead you tell it "Do the right thing 90% of the time, do something dumb 10%" and set it loose
2170[08:26:42.553] kanchess: tcecLC in clear #2 good chance at SuFi still
2171[08:26:49.238] gakuj505: next game pls Kappa
2172[08:26:50.626] lrmead: @xoto10 Not likely
2173[08:26:52.491] josephviruses: @xoto10 we've seen that in the past
2174[08:26:59.103] sometimesithurts: @L0RINDA hmm a joke? so what google's AUTOML is doing?
2175[08:27:03.678] lrmead: @L0RINDA Thus, human interference ...
2176[08:27:18.583] ジブ940 (jb940): Yes, the parameters are selected at the start of a training run. They're mostly inferred by checking how many positions are duplicates
2177[08:27:19.898] Aloril42: @L0RINDA Yes, might not yet be enough simple to avoid blunder.
2178[08:27:21.001] mariushav: @L0RINDA the joke might not be so silly! It's not entirely uncommon to use a sub-optimization routine for hyperparameters. Sometimes exotic stuff like evolutionary algorithms can work pretty well for that
2179[08:27:25.324] L0RINDA: @lrmead I mean it needs a program with which to function, they didnt sit it down in a room and have a chat :P
2180[08:27:27.454] noobs_killer: Leela blunder was a fairy tales these days.
2181[08:28:17.632] lrmead: @L0RINDA I am curious; why set such a parameter at all; let the machine decide ...
2182[08:28:20.189] josephviruses: Now leela is too solid, and very boring. It's most exciting game here was that against SF, imo
2183[08:28:24.771] Aloril42: Probably no blunder though, crash risk is higher.
2184[08:28:31.970] xoto10: leela not especially low on time yet though and game petering out ...
2185[08:28:37.971] bigdaas: sad sf is killing sf. sf will probably crash since the dll cannot be replaced
2186[08:28:55.745] Dabbler1: sf pawn is still on h1, obv didn't learn anything from this position Kappa
2187[08:29:11.479] mariushav: @lrmead in short, the training looks for the "lowest point in a terrain". If you just follow a path that is descending, you'll probably end up in a local low-point, but not a global one. By adding a "temperature", you introduce the option to randomly explore in "wrong" directions too
2188[08:29:13.563] sometimesithurts: @lrmead cuz they are hyper paramters not params
2189[08:29:15.336] xoto10: leela seems like sf in season 10
2190[08:29:16.046] ジブ940 (jb940): @lrmead because training takes months and months. To let the computer decide such parameters first of all needs many hundreds of such runs.
2191[08:29:17.901] chess_fan26: !dll
2192[08:29:18.101] Nightbot: ""Should Stockfish team be allowed to replace libwinpthread-1.dll with non-buggy version?" AntonMihailov: The final results of the voting are now in. Out of 8 votes we have 6 "yes", 1 "abstain", and one "no". This final no is the veto. Voting is anonymous and TCEC team will respect that and the outcome of the authors' vote nadedKing
2193[08:29:25.344] romanob: next 2 games interesting... sf could crash and leela could lose :-P
2194[08:29:33.749] Occyroexanthub: SF needs to be low on time in order to crash i think
2195[08:29:38.310] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2196[08:30:03.810] Occyroexanthub: as yesterday it did not respond for 7 minutes on a dead easy position but it played because it had too much time not to lose on time
2197[08:30:17.086] josephviruses: SF needs a short opening sequence to crash tcecK
2198[08:30:19.972] lrmead: @mariushav Like a system exploring the energy terrain; it may get stuck in a local minimum; quantum fluctuations can get it out ... right idea?
2199[08:30:24.229] Occyroexanthub: that "stall" was strange yesterday though
2200[08:30:25.412] xoto10: sf could crash, don't see leela losing with white
2201[08:30:25.848] josephviruses: Like a bookless opening tcecK
2202[08:31:03.195] mariushav: @lrmead to be honest, I have no clue about physics :)
2203[08:31:24.499] josephviruses: @xoto10 and leela might become undefeated again, but not promote to SuFi like SF did Kappa
2204[08:31:35.765] chess_fan26: So the one no wins over 6 yes
2205[08:31:37.362] fish_stock: SF should depart the tournament next game
2206[08:32:01.928] lrmead: @mariushav That, however, i precisely a good description for the evolution of a physical system which does not start in equilibrium ... it is the reason that Monte-Carlo methods can work.
2207[08:32:07.682] L0RINDA: @mariushav As I understand it (and I'm about the least technical person here), it is also what generates the slightly spurious looking evals in some situations. It has "lost" that dead drawn game in training due to temperature, so it knows its opponent can lose. (There's a case for saying that is also a form of contempt, as it can kind of predict the number of bad but playable moves in a position)
2208[08:32:42.326] lado933: a question: if an engine crashes 3 times i SuFi,does it just lose the 3 games or also gets disqualified and loses Sufi?
2209[08:32:44.917] tinduz_r: Yep. That
2210[08:32:57.823] mariushav: @lrmead that sound about right then!
2211[08:33:09.467] Occyroexanthub: no
2212[08:33:16.454] Occyroexanthub: just 3 losses
2213[08:33:20.254] Occyroexanthub: in sufi
2214[08:33:20.335] muppetmuppet: I think 3 crash rule still applies ?
2215[08:33:23.439] lrmead: Leela continues to fight here despite thinking it's down
2216[08:33:29.634] Occyroexanthub: 5 crashes=> 5 losses
2217[08:33:35.978] josephviruses: @lado933 only lose but not dq'd
2218[08:33:38.502] Occyroexanthub: x crashes => x losses
2219[08:33:40.331] Occyroexanthub: in sufi
2220[08:33:46.352] lado933: okay thanks
2221[08:33:46.625] xoto10: @josephviruses On the upside, sf became really good after that, so maybe leela will do the same :)
2222[08:33:50.396] muppetmuppet: oh ok sorry
2223[08:33:56.922] L0RINDA: @Occyroexanthub what about 7 crashes? :D
2224[08:34:02.733] mariushav: @L0RINDA yeah! In reality, even when the engine is self taught, there might be some bias in the parameter tubing and cross validation
2225[08:34:04.902] Occyroexanthub: 8 ? :P
2226[08:34:17.349] L0RINDA: :o
2227[08:34:18.630] trajkoskigoran02: Does this version of alliestein be released??
2228[08:34:31.767] chess_fan26: It has
2229[08:34:32.350] Occyroexanthub: it is already available
2230[08:34:38.073] Occyroexanthub: both eh binary and the net
2231[08:34:39.959] Aloril42: !download
2232[08:34:40.113] Nightbot: If you go to the !archive, under Seasons, each Season has its own little download icon to the right. And each event/division/league also has a download icon within each Season. - see also !downloads
2233[08:34:43.053] Occyroexanthub: *the
2234[08:34:46.220] Aloril42: !downloads
2235[08:34:46.419] Nightbot: S16 bonus: !ld2 Sufi openings S9-S15: http://blogchess2016.blogspot.com/2019/05/tcec-15-superfinal-opening-pgn-available.html !lczerodownload !stockfishdownload - see also !download
2236[08:34:47.293] josephviruses: @xoto10 hopefully. Contempt is not easy to train in leela. But who would know what the t60 nets might bring?
2237[08:34:47.793] halberder: @Occyroexanthub wait what? I thought everyone was saying 3 crashes = DQ yesterday
2238[08:34:58.754] muppetmuppet: I'm still waiting for a cpu version then it can play in muppet games
2239[08:34:59.398] Occyroexanthub: for divP
2240[08:35:01.988] Occyroexanthub: not for sufi
2241[08:35:17.667] halberder: ok
2242[08:35:35.848] mariushav: Has Allie been tested vs Lc0? How does it do?
2243[08:35:39.685] josephviruses: @muppetmuppet there is already ld2
2244[08:35:42.976] trajkoskigoran02: @ Occyroex Download link please??
2245[08:35:57.962] muppetmuppet: yes I have that
2246[08:35:59.683] Aloril42: Hmm.. what would be links for !alliesteindownload ?
2247[08:36:14.211] tinduz_r: !allie ?
2248[08:36:14.456] Nightbot: Allie is a neural network based engine written by Adam Treat, inspired by the AlphaZero paper and the Lc0 project. See - https://github.com/manyoso/allie , https://www.patreon.com/gonzochess75 and http://www.chessdom.com/alliestein-the-new-neural-network-entering-tcec-s15/ , Join the discord channel to follow Allie - https://discord.gg/UnbsJRU
2249[08:36:28.088] muppetmuppet: it isn't very good I claim the jh nets are better even on cpu
2250[08:36:29.591] Occyroexanthub: @trajkoskigoran02 https://drive.google.com/open?id=1KTv7chC4mL7LMEfUrYjve1NpkfhIDFZC
2251[08:36:31.154] Occyroexanthub: https://www.patreon.com/jjosh
2252[08:36:42.163] Occyroexanthub: 1st is binary. 2nd net
2253[08:37:00.939] Occyroexanthub: binary of TCEC
2254[08:37:07.892] josephviruses: @muppetmuppet PogChamp
2255[08:37:14.605] Occyroexanthub: for the net of TCEC S16 you have to pay 3-5 $
2256[08:37:51.784] trajkoskigoran02: @Occyroexanthub net link?? thank you.
2257[08:38:15.037] muppetmuppet: latest muppet game 30 in 300 mins still in opening just about French winawer r1bqrnk1/pp3ppp/2p2n2/3p2B1/1b1P4/3B2N1/PPP1NPPP/R2Q1RK1 w - - 4 11
2258[08:38:46.370] tinduz_r: Leela moobing?
2259[08:38:50.993] sirinial64: This DivP is pretty intense in controversies. SF crashing, dll fix vote and veto, allie is a clone, ... Cannot wait for the next controversy Kappa
2260[08:39:25.102] Occyroexanthub: Komodo MCTS is a clone of Komodo
2261[08:39:39.201] sirinial64: Duh LUL
2262[08:39:41.797] Occyroexanthub: maybe i should demand a vote? Kappa
2263[08:39:56.108] naw72: stoofvlees is really magnus carlsen incognito
2264[08:39:56.785] chess_fan26: Kmc is bugged this season
2265[08:39:58.874] Occyroexanthub: but they have almost the same name
2266[08:39:59.192] sometimesithurts: that is indeed a clone but no controversy there.. everyone agrees Kappa
2267[08:40:14.087] tinduz_r: Next controversy: we should only name engines after food. Kappa
2268[08:40:21.410] Aloril42: @Occyroexanthub @jjoshua2 Wasn't net available now for free? I remember seeing that in tcecLC discord..
2269[08:40:24.695] Archstro: Computer chess is a clone of human chess. This whole thing is a scam!
2270[08:40:25.798] sirinial64: @naw72 SeemsLegit
2271[08:40:47.448] jjoshua2: the latest 10.2 net is free as are most of them. but this one is still in early access
2272[08:41:02.960] CCLS: Houdini is a clone of...
2273[08:41:06.184] Occyroexanthub: 10.2 is free?
2274[08:41:10.556] Occyroexanthub: hmm i don't see it :P
2275[08:41:10.813] Nightbot: athHM
2276[08:41:24.477] trajkoskigoran02: @Occyroexanthub which net is used?? thank you.
2277[08:41:27.921] muppetmuppet: magnus Carlsen wld thrash these computers in a swimming pool
2278[08:41:32.891] jjoshua2: @Occyroexanthub it was never officially made a 10.2
2279[08:41:37.300] jjoshua2: its the one thats a SWA of 10.0 and 10.1
2280[08:41:39.437] Occyroexanthub: ask @jjoshua2 he is here :P
2281[08:41:45.965] jjoshua2: 10.0 is used here
2282[08:42:00.093] Occyroexanthub: which is not free
2283[08:42:01.344] jjoshua2: it scored slightly higher in a test than 10.1 and 10.2
2284[08:42:18.170] jjoshua2: technically they are all shortly free but early access for subscribers
2285[08:42:28.813] Occyroexanthub: well that's what i mean
2286[08:42:32.817] Occyroexanthub: free right NOW
2287[08:42:33.510] kbg519v1a: I notice a minor flaw in the interface. When it is Black to move and White has an eval in the range -0.08 to 0.08 but Black's eval is outside that range, then the Draw rule incorrectly says "9p". When it becomes White's turn to move then the Draw rule shows the correct "-".
2288[08:42:33.710] jjoshua2: its a perk for donaters
2289[08:42:36.510] Occyroexanthub: not in the future :P
2290[08:42:38.783] chess_fan26: @jjoshua2 the best net is 22 million steps, even better than 10.0
2291[08:42:45.133] sirinial64: 10.0 is in early access (I guess until divP ends?)
2292[08:42:59.430] dr_zugzwang: Does Leelenstein have T40 games?
2293[08:43:26.696] muppetmuppet: you mean alliestein ?
2294[08:43:39.925] dr_zugzwang: No, the net that allie uses.
2295[08:43:47.392] Occyroexanthub: he means Leelenstein :)
2296[08:44:00.885] Aloril42: !lczerodownload @jjoshua2 If it's somewhere available for download, can you create !alliesteindownload -command?
2297[08:44:01.101] Nightbot: https://github.com/LeelaChessZero/lc0/releases/download/v0.22.0/lc0-v0.22.0-windows-cuda.zip http://data.lczero.org/files/networks-contrib/T40B.4-160
2298[08:44:07.308] muppetmuppet: ok sorry I am not with it today at all
2299[08:44:19.174] Occyroexanthub: @Aloril42 it's not.
2300[08:44:21.904] Occyroexanthub: the TCEC net
2301[08:44:24.674] Occyroexanthub: it will be
2302[08:44:51.262] chess_fan26: @jjoshua2 am I right?
2303[08:44:57.321] Occyroexanthub: of course a command revealing the link right now is welcomed :P
2304[08:44:59.452] jjoshua2: 22 million steps is 10.0
2305[08:45:15.400] jjoshua2: alliestein uses the stein net
2306[08:45:20.855] dr_zugzwang: Why does Leelenstein have leela in its name ?
2307[08:45:37.233] Occyroexanthub: because it initially named it that way
2308[08:45:42.795] jjoshua2: here is the swa average of 10.0 and 10.1 https://www.patreon.com/file?h=28891632&i=4256072
2309[08:45:44.271] Occyroexanthub: as All were LEela.
2310[08:45:50.936] Occyroexanthub: training code etc
2311[08:45:53.585] lmabacus: @kbg519v1a That's not a flaw. Black's eval at that point is for a move that hasn't happened yet. There's always the possibility that black as eval drops into the draws one when the move is actually made.
2312[08:46:02.325] tinduz_r: Are there provisions in the licenses used by Leela that it should never be used for any proprietary gain? Also provisions on how its data/code should be cited for any use?
2313[08:46:03.171] jjoshua2: training code started out the same as leela's
2314[08:46:08.943] dr_zugzwang: Oh, so not merely a coindicence.
2315[08:46:12.785] jjoshua2: every new version of leelenstein has made an additional difference
2316[08:46:25.571] jjoshua2: it was originally mostly CCRL data
2317[08:46:25.611] bigdaas: gtg4now. cyall in the sufi!
2318[08:46:41.228] jjoshua2: !10.2
2319[08:46:49.101] chess_fan26: @jjoshua2 is the version playing here and at banana the same
2320[08:47:09.316] dr_zugzwang: Are Corchess, RubiFish etc. permitted in TCEC?
2321[08:47:16.403] Occyroexanthub: no
2322[08:47:23.347] chess_fan26: I mean net
2323[08:47:24.239] jjoshua2: !addcom !10.2 https://www.patreon.com/file?h=28891632&i=4256072 This is SWA average of stein 10.0 and 10.1. It needed a little reinforcement so is the base of the current training.
2324[08:47:24.582] Nightbot: @jjoshua2 -> The command "!10.2" has been added successfully.
2325[08:47:34.958] josephviruses: There are a lot of projects that originated from the leela projects to test algorithms that could not make within the zero framework
2326[08:47:35.375] jjoshua2: nets are the same
2327[08:47:42.285] dr_zugzwang: @Occyroexanthub Then why is Leelenstein?
2328[08:47:51.408] jjoshua2: allie doesnt have all the latest tuning though
2329[08:48:12.867] Occyroexanthub: Alliestein is permitted. i.e Allie binary + (Allie)stein net
2330[08:48:17.626] muppetmuppet: they have a 2 out of 3 policy
2331[08:48:18.005] josephviruses: LS was one of the earliest such projects iirc
2332[08:48:23.794] Occyroexanthub: oops got that wrong.
2333[08:48:49.704] Occyroexanthub: Allie-stein is permitted. that means Allie binary + (Leele)stein net
2334[08:48:56.765] chess_fan26: @jjoshua2 does stein have ICCF games?
2335[08:49:00.386] jjoshua2: gonzo was hard to reach beffore divp so he didnt get to test all my binary and tuning changes and suggestions
2336[08:49:05.238] Occyroexanthub: (*Leelen)stein
2337[08:49:05.607] enqwert81: @jjoshua2 Does AS have trade penalty??
2338[08:49:07.000] jjoshua2: @chess_fan26 no but i use that as validation
2339[08:49:08.848] sometimesithurts: missed an N
2340[08:49:43.579] jjoshua2: @enqwert81 thats up to gonzo in the binary, but the dataset of training games is a bit more aggressive with ccrl games it hink
2341[08:49:54.412] kbg519v1a: @Imabacus OK, I accept that.
2342[08:50:06.219] jjoshua2: @chess_fan26 its used both for tuning parameters and picking nets
2343[08:50:06.447] chess_fan26: @jjoshua2 then I think you should add those games
2344[08:50:13.713] josephviruses: Leela will run out of time soon.
2345[08:50:23.528] jjoshua2: if i were to add them to training they wouldnt be reliable for tuning parameters or picking which net was best
2346[08:50:41.616] jjoshua2: once these other things stop gainining elo it would be good for one last boost @chess_fan26
2347[08:50:44.632] muppetmuppet: old leela wld certrainly have a chance of losing this
2348[08:51:05.827] jjoshua2: see i have to change only one or two thigns at a time and have a easy comparator
2349[08:51:06.474] sirinial64: Training set /= validation set @chess_fan26
2350[08:51:08.969] jjoshua2: and then keep changes that work
2351[08:51:26.803] jjoshua2: gotta do yard work :(
2352[08:51:50.786] chess_fan26: @jjoshua2 the last boost idea is perfect
2353[08:52:14.154] tinduz_r: @jjoshua2 What's you opinion on KC saying Alliestein is not doing a good job giving credit to Leela in terms of the net?
2354[08:52:25.203] chess_fan26: @jjoshua2 I can't wait for stein to get T40 training data
2355[08:52:28.157] enqwert81: Allie gets more complicated positions compared to leela getting simplified endgames.
2356[08:52:34.702] jjoshua2: !blame @chess_fan26 for last 10 elo
2357[08:52:34.916] Nightbot: lupoNERD
2358[08:52:47.442] JISICHANCZ: How long time is it when SF did not qualify into the superfinal ??
2359[08:52:49.132] ジブ940 (jb940): Doesn't allie have trade penalty?
2360[08:52:55.590] jjoshua2: i really get to do some work so i cant fully address it
2361[08:52:59.726] enqwert81: It is more aggressive somehow.
2362[08:53:16.018] sirinial64: Thank you for dropping by @jjoshua2 . Alliestein is getting haters those days :( but you are doing a good job and you should be proud of stein's achievement in this divP
2363[08:53:31.612] muppetmuppet: I claim it is the net you can find old leela nets that are very agresive
2364[08:54:19.269] JISICHANCZ: NN era has already begun.
2365[08:54:19.524] Nightbot: drezzZZZ
2366[08:54:30.416] tinduz_r: !drawzone now
2367[08:54:30.632] Nightbot: Ladies and gentlemen, welcome to the draw zone [-0.08, 0.08] Pray for no captures or pawn moves BlessRNG
2368[08:54:36.898] muppetmuppet: I guess if the nets are made by playing against worse opponents in the old days they wld get more blunders and be more agressive
2369[08:54:42.055] jjoshua2: i agree we need a NN similiary test
2370[08:54:42.155] lado933: this game is not over yet. SF still can crash and leela still can blunder
2371[08:54:47.555] jjoshua2: and some new rules
2372[08:55:06.475] jjoshua2: like maybe engines have to score at least 10% differently on a test
2373[08:55:20.594] noobs_killer: !-
2374[08:55:20.785] Nightbot: NEG-A-TIVE
2375[08:55:22.295] jjoshua2: its very hard to do fair and design
2376[08:55:57.380] halberder: 10% differently? way too much. How much elo do you lose by replacing 10% of an engine's moves with the engine's second or third choices?
2377[08:56:00.820] ジブ940 (jb940): @jjoshua2 that's difficult, 90% of the AB engines would fail that. Even RF vs Leela. Leela has more moves in common with 0 contempt sf than 24 contempt sf does. Two Sf's in tourney?
2378[08:56:10.918] jjoshua2: well it has to be tuned for sure
2379[08:56:13.195] enqwert81: Tcec 2/3 NN rule feels OK to me
2380[08:56:30.994] jjoshua2: there can be additional datapoints at least even if its not the overall rule
2381[08:56:33.584] josephviruses: !-
2382[08:56:33.919] Nightbot: NEG-A-TIVE
2383[08:56:37.498] Twipply: I have LCZ's network, I find all positions where its second move is almost as good as its first move, I train my network to match those second moves n% of the time
2384[08:56:41.122] chess_fan26: @jjoshua2 if you try reducing the size of the stein, will it become weaker
2385[08:56:41.951] Twipply: Did I just circumvent your test?
2386[08:56:42.579] ジブ940 (jb940): I'm just saying the 0 contempt sf deviates more from 24 contempt sf than some other engines, literally.
2387[08:56:47.292] jjoshua2: something thats very different is obviously allowed and closer is required to submit to TD the lineage
2388[08:56:57.660] jjoshua2: i dont have all the solutions
2389[08:57:02.489] tinduz_r: To me personally, Alliestein is unique enough as a concept. Doesn't matter if one line of code is what makes it different. If one line makes Leela into an AB engine, then it's still unique.
2390[08:57:06.371] jjoshua2: @chess_fan26 stein used to be smaller
2391[08:57:08.911] jjoshua2: and it was much weaker then
2392[08:57:25.604] enqwert81: When engines are so strong, a lot of moves will be the same for sure.
2393[08:57:30.932] chess_fan26: @jjoshua2 okay thnx
2394[08:58:26.561] lrmead: !Negative
2395[08:58:42.128] halberder: how comparable are these engine ratings to FIDE ratings? no basis for comparison at all?
2396[08:58:45.913] lrmead: !negative
2397[08:58:56.177] jim_vs: !-
2398[08:58:56.499] Nightbot: NEG-A-TIVE
2399[08:59:06.111] noobs_killer: leela 3900, magnus 2800, ok?
2400[08:59:15.092] muppetmuppet: what do you mean compared to humans ?
2401[08:59:21.710] L0RINDA: @halberder The ratings here don't compare to FIDE.
2402[08:59:31.062] jim_vs: so Carlsen is Leela for me Kappa
2403[08:59:41.730] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
2404[08:59:51.013] halberder: they might compare to FIDE though if you used a very strong engine to compare average centipawn loss of other engines to human GMs
2405[08:59:58.755] jim_vs: @L0RINDA I think it does
2406[09:00:07.173] lrmead: @L0RINDA Why then are they here? I only understand Elo's for humans = win percent vs rating difference
2407[09:00:20.788] lrmead: Isn't it the same?
2408[09:00:21.080] jim_vs: !rating
2409[09:00:21.261] Nightbot: Elo, is it me you're looking for? --Lord_Aldrich_
2410[09:00:21.605] halberder: and then normed the ratings based on that
2411[09:00:22.595] L0RINDA: @lrmead to compare engines to other engines. The same system applies
2412[09:00:38.361] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2413[09:00:39.087] jim_vs: !ratings
2414[09:00:39.255] Nightbot: See Aloril's blog for information about TCEC ratings and monte carlo tournament calculations. https://tcecbayeselo.chessdom.org/ also ratings in GUI calculated with Ordo http://tcec-chess.com/ordo.txt
2415[09:00:40.064] ジブ940 (jb940): I just hope they do refixing of the ratings. Get the top 10 humans to play a bunch of games to engines rated around 2700. Fix those as baselines
2416[09:00:40.197] tinduz_r: !elo
2417[09:00:40.482] Nightbot: A typical Elo rating based on all previous TCEC games only. New engines get a rough estimate of starting Elo based on rating list or other info as decided by TCEC organizers. They are roughly equivalent to FIDE Elo scores, +/- 100 or so. Pedants disagree.
2418[09:00:45.454] lrmead: But his question was about humans vs machines ...
2419[09:00:46.337] muppetmuppet: are you saying if you take a ccrl 2800 engine it won't be about as good as Carlsen ?
2420[09:00:53.103] f1tof5: Sigh of relief as SF stays in the tournament.
2421[09:01:03.560] Twipply: @ジブ940 that'd make too much sense, sorry
2422[09:01:10.874] jim_vs: CCRL rating is not FIDE based. But TCEC is.
2423[09:01:18.972] L0RINDA: @muppetmuppet It'lll be somewhere in the ballpark
2424[09:01:33.361] xoroshiro: !sfprob
2425[09:01:33.709] Nightbot: see https://i.imgur.com/HUYlRze.png by lmabacus
2426[09:01:35.157] jim_vs: !moral
2427[09:01:35.408] Nightbot: Stephane_Nicolet: "One thing is sure, next version of Leela is the moral winner of TCEC 14 premium division"
2428[09:01:55.687] halberder: this is a draw by force I think
2429[09:01:57.906] chess_fan26: The voting had more 'yes' right
2430[09:01:59.696] muppetmuppet: ok thanks I still intend to challenge one of my d&d players in a chess match in the game sometime and will need about the right elo to make it a fair game
2431[09:02:02.896] jim_vs: Leela won the mini-match
2432[09:02:02.935] Aloril42: (script) 92: 3. tcecSF 3834(-1) - 2. tcecLC 3839(+1) tcecDD https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-92-stockfish-190826-lczero-v0-22-0-nt40b-4-160/
2433[09:02:03.909] alchemist888: @jim_vs TCEC is? Do't make me laugh :)
2434[09:02:04.718] lrmead: Probability wise then, 2800 Magnus has no chance vs Leelas 3900
2435[09:02:20.971] tinduz_r: No crash.
2436[09:02:44.074] jim_vs: I mean, TCEC team recalibrated ratings according to Kramnik rating
2437[09:02:50.814] L0RINDA: @muppetmuppet I suspect that lower ranked engines, especially if you can find the rating of an old out of service one, would be more accurate. They played humans more often back then
2438[09:02:54.797] jim_vs: !glaurung
2439[09:02:55.215] Nightbot: Glaurung is a UCI-compatible open source chess engine developed by Tord Romstad under the GPL 3. Glaurung is the forerunner of Stockfish by Marco Costalba, Joona Kiiski and Tord Romstad. First released in fall 2004, Glaurung is written in C, later versions completely in C++ - see https://www.chessprogramming.org/Glaurung
2440[09:03:06.613] jim_vs: !calibrate
2441[09:03:08.645] L0RINDA: @jim_vs Thanks, I hadn't realised they'd got that done
2442[09:03:16.579] jim_vs: !recalibrate
2443[09:03:18.823] muppetmuppet: ok yes might go back to a rebel I need about a BCF180 ish
2444[09:03:19.597] tinduz_r: !elo says it's roughly equivalent fo FIDE ratings +/- 100.
2445[09:03:19.835] Nightbot: A typical Elo rating based on all previous TCEC games only. New engines get a rough estimate of starting Elo based on rating list or other info as decided by TCEC organizers. They are roughly equivalent to FIDE Elo scores, +/- 100 or so. Pedants disagree.
2446[09:03:26.230] L0RINDA: so yeah, it's more representitive than I thought
2447[09:03:36.473] jim_vs: there is a command which explains that recalibrating, but I cannot recall it.
2448[09:03:43.459] chess_fan26: Sf has to not crash for 19 more games
2449[09:03:44.363] L0RINDA: @muppetmuppet looking at about 2050 then
2450[09:03:48.098] alchemist888: Did Kramnik play with some of TCEC engines?
2451[09:03:49.991] Aloril42: (script) if 1-0: Sco: 3717(+2), Sto: 3823(-11); if tcecDD : Sco: 3716(+1), Sto: 3831(-3); if 0-1: Sco: 3714(-1), Sto: 3841(+7); simulation and estimated start times: https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-93-speculation-simulation/
2452[09:03:52.795] muppetmuppet: right
2453[09:03:57.067] fish_stock: the partys over
2454[09:04:02.186] L0RINDA: yeah, i knew how it was going to work, just didnt realise it was done
2455[09:04:10.136] jim_vs: Kramnik played against Fritz
2456[09:04:10.254] gakuj505: pls win 1 black game, sf KappaWealth
2457[09:04:28.667] sirinial64: Leela has drawn sf in this French as black like a boss SeemsGood
2458[09:04:29.448] jim_vs: and a match was run Glaurung vs Fritz etc...
2459[09:04:31.756] L0RINDA: basically you can scale it up through various engines that have played other engines that have played other engines that have played humans
2460[09:04:31.997] fish_stock: its time to call it a day
2461[09:04:49.559] jim_vs: where is that command? :/
2462[09:05:02.297] sirinial64: !fritzinbahrain
2463[09:05:02.603] Nightbot: "Glaurung 2.2 x64(4 cores) - Fritz in Bahrain(1 core) +69 -10 =21 i.e, +235±72 Elo" -Occyroexanthub This is 20.5/100. See https://www.chessprogramming.org/Kramnik_versus_Deep_Fritz_2002
2464[09:05:09.312] fish_stock: goodbye stockfish we'll miss you
2465[09:05:43.426] jim_vs: !elorating
2466[09:05:47.436] L0RINDA: @muppetmuppet I know that lichess StockFish kinda cheats it with various levels. It's not quite a good fight though as it makes varying levels of forced mistakes rather than consistently playing a bit worse
2467[09:05:55.126] sometimesithurts: 0-1 and good bye leela
2468[09:05:58.353] L0RINDA: @muppetmuppet so the 1400 or whatever it is will still play great, but then hang a piece
2469[09:06:04.697] chess_fan26: Sf might save kmc from relegating
2470[09:06:07.770] muppetmuppet: yes I played that myself and found it not much fun
2471[09:06:21.999] L0RINDA: agree
2472[09:06:39.346] muppetmuppet: I got to about the 1900 one for myself
2473[09:06:45.748] fish_stock: 0-1 with a resounding crash
2474[09:06:49.976] alchemist888: But Kramnik have played only 6 games againts Deep Fritz 10 NotLikeThis
2475[09:06:54.850] ジブ940 (jb940): @L0RINDA what? That's exactly how I play. 2000 elo and 500 elo moves combined
2476[09:07:02.881] ジブ940 (jb940): Seems realistic Kappa
2477[09:07:22.783] L0RINDA: @ジブ940 it's not a mile from how i play tbf :D
2478[09:07:40.298] marklefler: @ジブ940 my system is easier. I just play 500 elo moves.
2479[09:07:53.354] muppetmuppet: I might try the bad gyal and evil gyal nets they are supposed to be more even
2480[09:07:58.930] sirinial64: Let's ask Kramnik to play 1000 games against Glaurung at long time control to get a better elo calibration Kappa
2481[09:08:03.327] Twipply: I just don't play at all
2482[09:08:10.866] tinduz_r: negative out of the book?
2483[09:08:22.981] Ove__: !gosco
2484[09:08:23.185] Nightbot: jaxiCrab
2485[09:08:31.691] ジブ940 (jb940): Stockfish is losing its mind.
2486[09:08:50.424] fish_stock: about move 20 i reckon
2487[09:09:09.784] sirinial64: !sconet the net training platform of Scorpio
2488[09:09:10.158] Nightbot: to donate GPU time for Scorpio training goto http://scorpiozero.ddns.net
2489[09:09:14.824] josephviruses: Leela agrees this is _slightly_ advantageous for black
2490[09:09:29.853] halberder: obviously if carlsen played this stockfish for 100 games he would get a few wins... because stockfish would crash
2491[09:09:38.951] alchemist888: CCRL list is close to FiDE rating than any other rating chart
2492[09:09:39.000] tinduz_r: bluela at 0.43 though
2493[09:09:40.038] ジブ940 (jb940): What? Leela is 0.43
2494[09:09:42.829] jim_vs: yeah, somebody should just pay Carlsen to play 100 games against Laser
2495[09:09:53.303] jim_vs: for recalibrating sake
2496[09:10:10.720] L0RINDA: !carlsen
2497[09:10:11.035] Nightbot: The TCEC-could-Carlsen-beat-this-engine Counter is now at 225 nadedKing
2498[09:10:19.492] sometimesithurts: against laser on a raspberry pie maybe Kappa
2499[09:10:27.672] sirinial64: Time control 40 moves in one minute Kappa
2500[09:10:37.885] jim_vs: or maybe play against the last engine of last league
2501[09:10:51.670] jim_vs: !l3
2502[09:10:56.863] L0RINDA: !l2
2503[09:10:57.846] jim_vs: !l4
2504[09:10:58.262] Nightbot: 1. Fritz= 2. Nirvana= 3. Arasan TCEC16* 4. Texel= 5. Vajolet= 6. Gull= 7. Pedone 230719* 8. Nemorino 5.17* 9. Rubichess 1.5* 10. Pirarucu 3.0.7* 11. ScorpioNN= 12. Stoofvlees II a11* 13. Marvin= 14. Wasp 3.69* 15. Topple= 16. chess22k 1.13* 11-16. Advanced from Qualification League. *:update, =:same
2505[09:10:59.131] sirinial64: !ql
2506[09:10:59.466] Nightbot: Qualification League for Season 16. New entries: 1. Stoofvlees II a10 2. ChessFighterNN 2.1 3. Asymptote 0.6.2 Updated Div4 engines: 4. Scorpio 2.9.1 Maddex 5.Winter 0.6, 6. Wasp 3.68 7. Bagatur 1.7b 8. Marvin 3.4.0 9. Topple 0.7.2 10. Cheese 2.1 11. Jumbo 0.6.117 12. Minic 0.76 13. Igel 1.8.1 14. Rodent III 0.287 Not updated: 15. Tucano 7.07 16. chess22k 1.12 (1 crash) 17. The Baron 3.44.1 Exhibit: 18. LCZero CPU, results not included for promotion.
2507[09:11:15.134] lrmead: Has anyone made a serious proposal to a high GM to play a match vs SF or Leela?
2508[09:11:19.160] jim_vs: against The Baron then
2509[09:11:25.155] L0RINDA: @lrmead There's no point
2510[09:11:32.024] sometimesithurts: why should one do that?
2511[09:11:32.079] sirinial64: I remember Jumbo finished last of QL
2512[09:11:32.958] L0RINDA: or Igel from last season
2513[09:11:46.043] tinduz_r: Does anybody have an estimate how far ahead in moves do these engines look ahead based on nodes or nps?
2514[09:11:52.869] jim_vs: @L0RINDA there is a point to recalibrate Elo Kappa
2515[09:11:59.517] L0RINDA: :D
2516[09:12:01.083] halberder: what if the human got extra time
2517[09:12:03.125] muppetmuppet: they have a depth
2518[09:12:03.772] lrmead: @L0RINDA It would be extremely interesting match ...
2519[09:12:15.906] Ove__: !db
2520[09:12:17.275] eval_bot (10w) +4=11-7 • Bd3 (+3=6-4) • Be2 (+0=4-3) • g3 (+1=1-0) • [1997] Aleksandrov, Aleksej (2660) vs Ehlvest, Jaan (2610): draw • [1997] Chiburdanidze, Maia (2525) vs Adams, Michael (2680): black • <Lichess>
2521[09:12:21.752] L0RINDA: @lrmead It really wouldnt, Carlsen would be exterminated brutally
2522[09:12:24.975] tinduz_r: Oh yeah... missed that. Depth.
2523[09:12:26.945] sometimesithurts: Jumbo on a raspberry pie 2 seems fair
2524[09:12:31.850] halberder: carlsen @ 1 month/move vs leela at 5 min/move without pondering.
2525[09:12:44.652] Twipply: Might be an interesting match from the chess engine perspective, but Carlsen/etc would get violated
2526[09:12:48.090] Twipply: not much fun for them
2527[09:12:50.735] lrmead: @L0RINDA Sure? Anyway, still be very fun to see
2528[09:12:51.385] Archstro: !carlsen
2529[09:12:51.636] Nightbot: The TCEC-could-Carlsen-beat-this-engine Counter is now at 226 nadedKing
2530[09:12:53.321] L0RINDA: @lrmead He would need to play one of the very bottom engines to be able to score wins
2531[09:13:03.607] jim_vs: @L0RINDA all we ask is that Carlsen draw some, to be able to calculate Elo performance!
2532[09:13:06.977] chess_fan26: @Ove__ why didn't igel play this season
2533[09:13:32.425] Twipply: @L0RINDA I don't believe you're meant to use elo when the rating difference is over about 400 anyway
2534[09:13:40.311] jim_vs: 0.5/100 games is sufficient
2535[09:13:41.905] Ove__: Igel 1.8.1 played this season. There is also a major Igel update for next season
2536[09:13:46.414] sirinial64: One month a move = 10 years to finish ONE 120-move game Kappa
2537[09:13:48.480] lrmead: @L0RINDA I would even take Anand or someone else in the top 10
2538[09:13:51.064] tinduz_r: How about having Carlsen start from QL? Kappa
2539[09:13:52.334] sometimesithurts: !elo
2540[09:13:52.493] Nightbot: A typical Elo rating based on all previous TCEC games only. New engines get a rough estimate of starting Elo based on rating list or other info as decided by TCEC organizers. They are roughly equivalent to FIDE Elo scores, +/- 100 or so. Pedants disagree.
2541[09:14:05.689] L0RINDA: @jim_vs Joking aside, maybe a 2500 could be paid to have a go against an older engine
2542[09:14:16.944] L0RINDA: Where the money might be meaningful to them
2543[09:14:19.261] chess_fan26: @Ove__ thnx
2544[09:14:20.844] lrmead: What are Pedants?
2545[09:14:22.642] sometimesithurts: what was the link for tcec elo list?
2546[09:14:32.964] noobpwnftw: is @alyant around?
2547[09:14:34.305] Ove__: @chess_fan26 a +100 Elo update
2548[09:14:36.399] Archstro: An 2500 human could be playing like a 2300 or 2700 on a particular day. How are you going to account for that?
2549[09:14:42.880] muppetmuppet: the are people who are pedantic
2550[09:14:43.941] jim_vs: @L0RINDA yes, that's the point
2551[09:14:57.871] ジブ940 (jb940): @Archstro do it a week long.
2552[09:15:00.307] L0RINDA: @Twipply correct, 400 is the magic number (roughly)
2553[09:15:00.841] Twipply: @Archstro play many games over a period of time so any ups and downs average out
2554[09:15:11.184] chess_fan26: @Ove__ I guess Igel will promote this season to l2
2555[09:15:27.613] Ove__: SeemsGood
2556[09:15:27.820] sirinial64: @Archstro repeat the match 100 times and take the average score Kappa
2557[09:15:37.351] L0RINDA: @Twipply you'd run a 2500 vs a 2500ish machine and extrapolate through results working up, there's zero point playing against Carlsen, hence jim's Kappa :D
2558[09:15:37.504] Nuclearball: What is igel? New engine?
2559[09:15:45.166] muppetmuppet: you might not be able to use a single human they might have an anti computer style or a terrible style to play computers
2560[09:15:52.278] L0RINDA: also true
2561[09:16:02.501] Ove__: !igel
2562[09:16:02.700] Nightbot: Igel, a derivative of GreKo 2018.01, is written in C++ by Volodymyr Shcherbyna. Lazy SMP was added in the initial release January 2018. See https://www.chessprogramming.org/Igel
2563[09:16:02.733] L0RINDA: plus some humans just play badly vs engines
2564[09:16:21.203] Nuclearball: Ty @Ove__
2565[09:16:27.406] sirinial64: !reverse
2566[09:16:27.579] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
2567[09:16:30.610] eval_bot Game #65 "Stockfish 190826" vs. "ScorpioNN v3.0.1-n_maddex_INT8": [book] 9. Nf4 Nc6 10. Be2 Bb7 11. O-O g6 12. b4 Bf6 13. Rab1 Ne7 14. Rfd1 a5 15. b5 Rb8 16. Qc2 Bg7 17. a4 c5 18. dxc5 bxc5 19. Ne4 Bxe4 20. Qxe4 Nc8 21. Qc2 Nb6 22. g3 Qf6 23. Qa2 Rfd8 ... 1-0
2568[09:17:06.379] L0RINDA: I have an old chess computer that's 135 ECF, we could start there :P
2569[09:17:42.647] L0RINDA: Maybe KC would be up for it :D
2570[09:17:47.774] L0RINDA: get him some YT content
2571[09:17:48.839] anshul_mangla: come on SFFFF! come on!
2572[09:18:03.201] L0RINDA: then we can start over 2000 at least
2573[09:18:43.043] tinduz_r: If KC draws against an engine, I would be mildly surprised.
2574[09:18:53.171] L0RINDA: @muppetmuppet Which region do you play in? I'm playing in Lincs league this year, first time in 25 years or so
2575[09:19:11.536] muppetmuppet: I pretty much gave up chess 30 years ago
2576[09:19:22.001] L0RINDA: @tinduz_r depends on the engine, we're not talking a current TCEC engine
2577[09:19:25.538] AfuroZamurai: its easy to beat/draw an engine, just take the right and weak one Kappa
2578[09:19:28.578] jim_vs: maybe get Carlsen estimate Elo of a 2500ish Engine and then extrapolate
2579[09:19:45.307] fish_stock: come on ladies come on ladies £1 feeeesh... £1 feeesh
2580[09:19:46.639] muppetmuppet: my friend is in the bham leaguehe has won the individual about 7 times
2581[09:19:59.796] tinduz_r: There are also engines that seem to have exploitable "bugs" you could say.
2582[09:20:06.195] L0RINDA: Awesome. I used to play in the West mids league, but that was 1991 or so :D
2583[09:20:34.538] tinduz_r: Like the one Nakamura beat due to not wanting 50 move rule to happen.
2584[09:20:38.599] muppetmuppet: I played from about 76 to 84 ish
2585[09:20:40.353] L0RINDA: and I won't have played any of you because I was bad at chess, and still am :P
2586[09:20:52.855] L0RINDA: I may have played in Oxon in 84
2587[09:21:16.626] L0RINDA: Played board 20 as a junior when they were a man short :D
2588[09:21:28.008] muppetmuppet: I play bridge these days world champs start tomorrow no computers yet
2589[09:21:59.978] Twipply: No bridge engines at all?
2590[09:22:02.043] L0RINDA: I was put off bridge by virtue of having to explain to the opponents what my bidding meant. Silly I know, but was something I could never get over as a rule
2591[09:22:13.875] muppetmuppet: I mean ones that beat the best humans sorry
2592[09:22:25.424] L0RINDA: I wanted to design a great bidding system that I could keep secret with my partner
2593[09:22:25.590] Twipply: Oh. Okay
2594[09:22:43.921] L0RINDA: I do understand all the flaws in that reasoning
2595[09:22:54.691] muppetmuppet: they are close in 2005 one narrowly lost to a top dutch side yes unfortunately bidding systems have to be explained to the opponents
2596[09:22:57.991] Twipply: I would like to see some card game engines, like Magic The Gathering or Hearthstone
2597[09:23:12.334] L0RINDA: HS has had bots from time to time
2598[09:23:28.946] nickpelling: @L0RINDA that's not really how bidding works. :-(
2599[09:23:29.401] L0RINDA: games of incomplete information are tough though, but yeah will be fun when it happens
2600[09:23:35.673] L0RINDA: @nickpelling I know :(
2601[09:23:37.903] muppetmuppet: yes not seen a magic engine that is anything but terrible
2602[09:23:41.283] L0RINDA: @nickpelling It's what I wanted to do though :D
2603[09:23:42.382] Twipply: I suppose the in game bots already count in a way
2604[09:23:47.530] Twipply: I mean actual tournaments and stuff though
2605[09:23:51.299] Twipply: Fun to see perhaps
2606[09:24:06.530] L0RINDA: Yeah, I'd have fun casting that :D
2607[09:24:08.101] Nuclearball: That 5 pawn formation LUL
2608[09:24:09.785] Twipply: I considered it but no way am I writing a Hearthstone library to build an engine off
2609[09:24:45.562] muppetmuppet: there are no NN bridge engines yet but they are being looked at
2610[09:25:04.249] drfudgeboy: !when
2611[09:25:04.558] Nightbot: Premier Division started at https://tcec-chess.com 11 days 22 hours 25 minutes 34 seconds ago
2612[09:25:34.881] drfudgeboy: !time
2613[09:25:36.242] eval_bot 76/168 games left • Average duration: 02:59:02 • Estimated division end: Tue, 24 Sep 2019 01:07:17 GMT • Shortest game: 00:59:55 • Longest game: 04:06:54
2614[09:25:45.498] nickpelling: Redfish expecting a full-blown pawn storm, blackfish expecting... tea and cakes with the vicar.
2615[09:25:46.544] muppetmuppet: they did beat starcraft though so presumably it is time for a new target
2616[09:25:57.251] L0RINDA: They didnt REALLY beat SC though :P
2617[09:26:04.078] L0RINDA: It was pretty dubious
2618[09:26:28.595] muppetmuppet: ok the burst micro was maybe dodgy true
2619[09:26:31.509] L0RINDA: They beat the micro, but part of that is because the computer had perfect control (even though it was slowed i believe to try to be fair)
2620[09:26:50.320] L0RINDA: I'd like to see NNs take on something like Civ against good players
2621[09:27:04.826] Twipply: Even if they did win a fair match, it'd be like the Kasparov vs Deep Blue thing
2622[09:27:11.316] L0RINDA: That's kinda a way of neutralising the mechanical aspect
2623[09:27:17.083] Twipply: Given a little time and data to analyse, humans would come up with new strategies against the engine
2624[09:27:33.820] Twipply: New openings in SC come out all the time and take a while to be figured out
2625[09:27:41.188] sirinial64: I think poker texas hold'em no limit still resists to AI players
2626[09:27:47.718] gakuj505: Kappa
2627[09:27:52.933] L0RINDA: full table yes, not so sure about heads up
2628[09:27:53.337] muppetmuppet: heads up is done
2629[09:28:32.605] gakuj505: a6 b6 c6 d6 e6 h6 --> f6 g6f6 g6 Kappa KappaRoss Keepo KappaClaus DarkMode
2630[09:28:45.031] nickpelling: AI thumb war is vicious.
2631[09:29:39.255] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2632[09:29:54.512] sirinial64: 13. .... Nd7 first non-pawn move of sf PogChamp
2633[09:31:03.521] alayant: 800 elo pawn structure by SF ? :P
2634[09:31:25.221] nickpelling: Pieces are overrated. Underdevelopment is the new new thing.
2635[09:31:25.596] noobpwnftw: @alayant new book is ready
2636[09:31:29.757] noobpwnftw: https://github.com/official-stockfish/Stockfish/issues/2283#issuecomment-531478866
2637[09:31:58.143] nickpelling: !-
2638[09:31:58.338] Nightbot: NEG-A-TIVE
2639[09:32:06.569] WilliamLP: So, is this the beginning of the AllieStein era?
2640[09:32:28.260] muppetmuppet: no
2641[09:32:47.040] saulzar: !+
2642[09:32:47.283] Nightbot: POS-I-TIVE
2643[09:32:54.017] xoto10: while white go for c5 and the TheIlluminati formation? :)
2644[09:32:57.912] snakesanders22: allie is 1st?? what happned
2645[09:33:06.057] xoto10: *will* white ...
2646[09:33:22.195] WilliamLP: End of an era for Leela?
2647[09:33:59.606] alayant: @noobpwnftw Oh, nice. 150K positions for the 3 moves one, I guess the 4 moves one is over a million ?
2648[09:35:18.435] noobpwnftw: 1837365
2649[09:35:34.439] alayant: I see you decided against removing lines with an eval close to 0.00 for the sake of keeping diversity.
2650[09:35:42.756] noobpwnftw: yes
2651[09:36:12.825] WilliamLP: Stockfish is finally losing its edge in tournaments, when playing more NN engines. It used to rack up points by thrashing inferior AB engines, and there are fewer of them now.
2652[09:36:28.941] muppetmuppet: what did you do chop -1 tand below and +1 and above ?
2653[09:37:16.154] noobpwnftw: pretty much
2654[09:37:46.055] Vizvezdenec: I don't think you need to chop -1 and +1 tbh
2655[09:37:55.766] alayant: The lines with an eval close to 0.00 but with a complex position are good as it's still hard to find the right moves ; but those where it's relatively easy to find moves leading to a draw are not. This is not so easy to separate because of the nature of your book, it goes very deep in some lines and not so much in others
2656[09:37:56.138] Vizvezdenec: there are a lot of +1 which sf can't win
2657[09:38:19.486] jim_vs: if not for the crashes, SF wouls have 1 point more, that's right?
2658[09:38:22.466] muppetmuppet: I am surprised I am still getting reasonable opening diversity with no opening book
2659[09:38:52.833] noobpwnftw: I did perft 4 or 5 IIRC so it doesn't matter
2660[09:39:20.820] muppetmuppet: but I only play one game a day ish
2661[09:39:21.683] noobpwnftw: I evaluated every single move for at least 4 moves from startpos for depth 20
2662[09:39:26.843] noobpwnftw: or 22
2663[09:40:31.214] muppetmuppet: currently getting lots of French winawer's
2664[09:41:25.326] alayant: I'd like to see a book generated with somewhat different methods from chessdb, for the sake of comparing it with those. The potential sensitivity gains/diversity losses are rather abstract rn
2665[09:42:32.268] kanchess: !e44
2666[09:42:32.477] Nightbot: E44 Nimzo-Indian, Fischer Variation, 5.Ne2
2667[09:42:56.269] kanchess: tcecSF in Fisher variation PogChamp
2668[09:42:59.535] creyzamo: SF playing three-letter-man system LUL
2669[09:43:37.645] noobpwnftw: my experience told me they are irrelevant
2670[09:44:06.831] noobpwnftw: I can build chessdb with fruit or crafty and still give you the same conclusion, just a lot slower
2671[09:44:25.146] alayant: I mean a fishtest opening book
2672[09:44:38.768] alayant: Of course chessdb could be built with other engines too, considering its method
2673[09:44:49.445] noobpwnftw: there is some truth in the statistics
2674[09:45:15.679] noobpwnftw: like leela, its conclusion fits statistics more than anything else due to its nature
2675[09:45:38.594] alayant: It could go something like this : at each ply, cut out moves that lose more than 0.5 or so (so on start pos, no g4, f3, Na). Once you get all 6-pliers that can be reached this way, cut out lines that are over +1.2 or between -0.2 and +0.2.
2676[09:45:42.191] noobpwnftw: correct or not, it is irrelevant for a few moves from startpos anyway
2677[09:45:53.234] kanchess: !define test
2678[09:45:53.447] Nightbot: goto https://en.wiktionary.org/wiki/test nadedKing
2679[09:46:04.944] kanchess: ah works
2680[09:47:10.506] noobpwnftw: I did that already @alayant
2681[09:47:40.134] noobpwnftw: my PHP script is arcane, containing many unexplored features including the one you just said
2682[09:47:49.247] noobpwnftw: Kappa
2683[09:48:44.466] noobpwnftw: except I didn't cut lines around draws that's all
2684[09:48:44.725] kanchess: !-
2685[09:48:44.914] Nightbot: NEG-A-TIVE
2686[09:50:20.690] skembris: go f6 next time..
2687[09:50:23.953] alayant: Your PHP script is indeed arcane :P
2688[09:50:25.007] muppetmuppet: just f6 to play now
2689[09:50:38.511] gakuj505: f6 PogChamp PogChamp
2690[09:51:09.607] noobpwnftw: check that throttle thing
2691[09:51:45.745] noobpwnftw: it will cut nodes that is outside a window of best score, then by the effect of ply counter, those will not be in the output
2692[09:52:38.996] alayant: I'd like to see what it does with lines cut around draws - I prefer having things tested out than just relying on impressions.
2693[09:53:28.186] sometimesithurts: !cdb
2694[09:53:29.560] chessdbcn (21w) • Bf3 12 ! (00-00) • <ChessDBCN>
2695[09:53:34.945] noobpwnftw: simple, probe them from my API(queryscore) with a program
2696[09:53:42.363] muppetmuppet: you cld use human games to decide whether to cut lines, kill ones with more than 50% draws maybe ?
2697[09:54:24.810] noobs_killer: !-
2698[09:54:25.364] Nightbot: NEG-A-TIVE
2699[09:54:32.485] alayant: Human games :D
2700[09:54:43.864] Hanamuke: nice pawn structure
2701[09:54:55.731] noobs_killer: !sfcrash
2702[09:55:03.525] noobpwnftw: that looks 800 elo visually
2703[09:55:04.613] noobpwnftw: Kappa
2704[09:55:12.174] muppetmuppet: well I was assuming there wasn't a big enough db of machine games and you wld have to kill lines with over 80% draws then I guess
2705[09:55:35.411] alayant: We are not using a db of games
2706[09:55:41.456] alayant: But a db of positions
2707[09:55:46.075] alayant: !cdb
2708[09:55:47.177] chessdbcn (21b) • Nf6 -24 ! (00-00) • <ChessDBCN>
2709[09:55:47.630] fishtrawler: Are we out of book now?
2710[09:55:50.444] muppetmuppet: oh I see
2711[09:55:58.291] gakuj505: KappaPride
2712[09:56:23.042] oci82: why not f6 pff
2713[09:57:15.002] alayant: !cdb is never out of book at TCEC Kappa
2714[09:57:24.474] gakuj505: f5 PogChamp PogChamp
2715[09:57:35.572] chessdbcn (22b) • bxc5 10 ! (00-00) • <ChessDBCN>
2716[09:57:51.649] fishtrawler: ok
2717[09:57:54.503] gakuj505: *c5
2718[09:58:13.057] xoto10: haha, spear formation vs the line LUL
2719[09:58:14.637] alayant: I don't think any other game ever reached this position however :p
2720[09:58:39.177] noobpwnftw: battle of formations!
2721[09:58:40.092] xoto10: we just need to see a roman phalanx now :)
2722[09:58:52.631] muppetmuppet: or a square
2723[09:59:18.183] xoto10: I guess a 2x2 square of pawns is possible
2724[09:59:23.707] muppetmuppet: we did have a game a while back with a piece on every square of the e or d file can't remember which
2725[09:59:28.039] noobpwnftw: white's noob gun formation is nice
2726[09:59:30.135] xoto10: 3x3 seems unlikely :P
2727[09:59:31.647] noobpwnftw: look at that
2728[09:59:43.124] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
2729[09:59:55.736] xoto10: and black has one too
2730[10:00:08.505] xoto10: right where the pawns might open up as well
2731[10:00:15.045] sometimesithurts: wow finally the TheIlluminati
2732[10:00:58.116] gakuj505: ResidentSleeper
2733[10:00:59.265] sometimesithurts: white did played c5 ! coincidence??
2734[10:01:02.634] Occyroexanthub: SF like Kasparov (see till move 14) http://www.chessgames.com/perl/chessgame?gid=1070912
2735[10:01:07.259] Occyroexanthub: i.e a wuss Kappa
2736[10:01:14.842] fish_stock: OMG get ready
2737[10:01:15.086] Nightbot: Oh My Gull niff1
2738[10:01:34.606] noobpwnftw: yeah stop there for the rest of the hour Kappa
2739[10:01:39.892] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2740[10:01:57.025] noobpwnftw: then this game must be legendary
2741[10:02:16.320] flitzekacke: I love that pawn structure
2742[10:02:17.054] xoto10: where's @Vizvezdenec ? we need that nearlyUnwinnable patch, sf going for pawns on ks ending :(
2743[10:02:22.670] noobpwnftw: a century later it will be on history's mysteries
2744[10:02:48.080] muppetmuppet: queston of sport what happened next
2745[10:02:53.452] noobpwnftw: people argue which side has a win Kappa
2746[10:09:05.169] sometimesithurts: a century later chess is solved by cdb Kappa
2747[10:09:35.982] sometimesithurts: then people only need to look at that, not arguing LUL
2748[10:10:10.311] Vizvezdenec: going for pawns where?
2749[10:10:25.098] noname8365: if chess is solved people will argue which lines are sharper than others
2750[10:11:14.341] sometimesithurts: not if the position is a forced win :)
2751[10:11:24.339] noobpwnftw: I think cdb is one of those things that the goal can never be reached, just infinitely close Kappa
2752[10:11:58.550] kanchess: TB 31.9999999999 in fact
2753[10:12:44.374] xoto10: oh, changed now. It had an ending with a couple of pieces and fgh pawns for each side. BORING!
2754[10:13:23.690] Vizvezdenec: I guess this is why sf was showing 0.00 Kappa
2755[10:13:27.450] halberder: to truly solve chess we would need to start proving general theorems about who can win from a position
2756[10:13:32.928] xoto10: although I guess it's at 0.00 just now, so even your patch would be happy with that ending
2757[10:13:41.862] xoto10: yes :)
2758[10:13:48.830] Vizvezdenec: it will still show 0.00
2759[10:13:54.151] Vizvezdenec: even deeper 0.00 Kappa
2760[10:14:08.378] sometimesithurts: but the position indeed was eye-gouging (esp if f7 f6 swapped) TheIlluminati vs Communist : 1rr2b1k/1bq2p2/pppppnpp/2P5/1P1P1NB1/P1N1P2P/2Q2PP1/2RR2K1 b - - 0 22
2761[10:14:11.912] LoIiSquad: 0.000?! PogChamp
2762[10:15:08.123] Aloril42: @noobpwnftw Infinite with finite amount of positions. tcecK
2763[10:15:13.697] halberder: for example, a good start would be, if one side has a massive material advantage, determining what are the "exception" positions where the other side wins, so that if the current position is not an exception, the side with material advantage must win
2764[10:15:44.523] halberder: e.g. "player A is up two queens, and player B does not have any immediate mating threats or promotion threats within X moves, therefore player A must win."
2765[10:15:49.880] Aloril42: !tb32 @halberder Or just build enough big computer ;)
2766[10:15:50.078] Nightbot: Upper limit of positions reachable from starting position is <10^47 https://en.wikipedia.org/wiki/Chess#Combinatorics_of_chess_and_chess_puzzles Avogadro constant is ~6*10^23 mol^-1 and mass of Saturn is ~6*10^26 kg Question is not about number of atoms, but whether they can be organized so that they generate 32-piece TB for positions reachable from starting position. Solving single position is however potentially MUCH easier than generating 32-piece tcecTB . See !ab
2767[10:16:15.394] halberder: yeah that's not gonna happen :)
2768[10:16:25.094] xoto10: !-
2769[10:16:25.351] Nightbot: NEG-A-TIVE
2770[10:16:49.285] Twipply: Sounds a bit like futility pruning there
2771[10:16:57.490] sometimesithurts: who knows, a century later maybe noob has 1000x quntum gpu running cdb
2772[10:16:58.503] Twipply: "I'm so far ahead you can't catch up"
2773[10:17:15.911] halberder: it's different though because here we're talking about a *proof* that the other player can't catch up, not just a heuristic guess
2774[10:19:21.350] nsousa254: SCAM
2775[10:19:25.025] nsousa254: you heard it first here
2776[10:19:54.726] halberder: ?
2777[10:19:54.960] Nightbot: lirikTHINK
2778[10:20:14.518] Vizvezdenec: xoto you need to become an approver btw
2779[10:20:23.242] polkiujmn: SF negative PogChamp
2780[10:20:31.171] sometimesithurts: but if TB32 happens when I'm at my 80s, I feel like my whole life is ruined.. spending it on computer chess
2781[10:20:53.816] polkiujmn: lol you'd be lucky get get tb10
2782[10:21:07.871] Vizvezdenec: @xoto10 are you OK with it?
2783[10:21:08.594] halberder: to solve chess would require working not with individual positions, but with *classes* of positions that all have similar properties
2784[10:21:24.362] polkiujmn: honestly i just don't think anyones reallythat motivated to brute force tbs
2785[10:22:30.862] polkiujmn: like maybe if theres a major breakthrough in computing technologies and it becomes relatively easy to compute tbs then someone would probly do it
2786[10:22:49.707] halberder: to take a simple example - in the endgame, in KPk endings, we can concisely summarize which positions result in a win by reference to key squares, *without* needing to store every individual position and its result
2787[10:23:39.159] xoto10: no, not sure that's a good idea!
2788[10:23:45.922] Aloril42: @halberder Solving all 32 piece positions or just standard starting position?
2789[10:23:59.838] Vizvezdenec: why DansGame
2790[10:24:14.852] halberder: I don't think either would ever happen by brute force @Aloril42
2791[10:25:02.381] sometimesithurts: 32 piece position is cooler since we would already solve chess960 as well Kappa
2792[10:25:13.608] Aloril42: @halberder Later one might be just square root of all positions or about 10^24 (or less)
2793[10:26:14.865] halberder: why the square root?
2794[10:26:23.835] Aloril42: !ab
2795[10:26:24.071] Nightbot: AB = Alpha-Beta. The Alpha-Beta algorithm (Alpha-Beta Pruning, Alpha-Beta Heuristic) is a significant enhancement to the minimax search algorithm that eliminates the need to search large portions of the game tree applying a branch-and-bound technique. See https://www.chessprogramming.org/Alpha-Beta
2796[10:27:27.777] sometimesithurts: how this is related to square?
2797[10:28:02.138] Aloril42: @sometimesithurts Optimal AB is square root of minmax.
2798[10:28:08.506] lrmead: Ok Scorpio, keep it up ...
2799[10:28:17.281] noname8365: only with optimal move ordering from my understanding though
2800[10:28:23.068] sometimesithurts: for a weak solve, we need to search all the possible moves for black
2801[10:29:13.358] sometimesithurts: AB might start to evaluate a position which is evaluated by black as -12
2802[10:29:40.352] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2803[10:29:43.003] halberder: for a complete solve, all evaluations have to be either "W" "D" or "L", no numeric guesses
2804[10:29:48.945] sometimesithurts: *stop to evaluate (wow 30 sec wait :/ )
2805[10:31:22.758] sometimesithurts: of course it is almost unlikely for black to play it (cuz if it does it will be crushed sooner) but the the engine should continue to evaluate all possible plays by black to reach a check mate
2806[10:31:56.027] L0RINDA: !time 94
2807[10:31:57.422] eval_bot Game #94 "LCZero v0.22.0-nT40B.4-160" vs "Komodo 2381.00" is estimated to start on Sat, 14 Sep 2019 16:04:27 GMT (After 1h 32')
2808[10:32:32.991] L0RINDA: Could be looking at an Allie - Komodo final in some situations, that's a phrase I wasn't expecting to say anytime soon
2809[10:32:43.126] noobs_killer: !NN
2810[10:32:43.448] Nightbot: NN means Neural Network. NN programs do a huge number of inner products. A gpu and similar processors include circuitry for very fast vector inner product (dot product). Therefore, NN based engines gain from access to a gpu. Non NN engines do not have a (known) way to take advantage of gpus.
2811[10:35:00.164] sometimesithurts: @noobs_killer ops not a good ID here.. noob's got a lot of fans in tcec chat Kappa
2812[10:36:24.961] sometimesithurts: unless it is actually noob's killer LUL
2813[10:38:39.082] halberder: if there really were only 10^24 positions needed then that might be feasible to brute force, but I am doubtful.
2814[10:39:05.656] nonsio: !endofera
2815[10:39:05.867] Nightbot: tcecNE 22 years 4 months 2 days 14 hours 39 minutes 36 seconds ago since end of human era nadedKing
2816[10:39:08.092] fire2fire: brute force with finger
2817[10:39:35.710] nonsio: !newera
2818[10:39:36.003] Nightbot: achesBG
2819[10:40:33.825] fish_stock: what if this vote thing was a ruse and tcec has surreptitiously replaced the binary for SF already CoolStoryBob
2820[10:42:02.186] gakuj505: fish can't win while play black game NotLikeThis
2821[10:42:19.846] halberder: it might be stored atom by atom, but it would require a huge computing effort for one thing, as 10^24 nanoseconds is 30 million years, assuming one position per nanosecond per processor
2822[10:43:54.873] halberder: the human race would have to be really bored to devote so much effort
2823[10:45:11.663] halberder: I think it's more practical to try to build w/l/d tables for whole *classes* of positions, where each class of position is concisely summarizable and every member of the class has the same w/l/d outcome
2824[10:47:12.009] Nuclearball: Scorpion crash?
2825[10:47:38.516] halberder: it's contagious!
2826[10:48:32.722] polkiujmn: lol nah
2827[10:48:47.109] nsousa254: how's the scam going?
2828[10:49:01.170] naw72: waiting for SF to crash
2829[10:49:11.450] nsousa254: good. all according to the MASTER PLAN
2830[10:49:37.195] Aloril42: @halberder So have billion processors and do it in less than year :)
2831[10:50:26.334] Aloril42: @halberder Now using match to solve is no doubt better idea, just saying brute force might also be feasible.
2832[10:51:17.404] naw72: with a quantum computer you could win a game before starting it
2833[10:52:02.152] leotrubach2: poor black pawns
2834[10:52:48.454] lado933: this looks like a position where NNs can blunder
2835[10:53:10.858] leotrubach2: I think white is pretty safe here
2836[10:53:41.441] lado933: scorpio could become too confident and go for the win
2837[10:56:18.020] chetti95: It annoys me to see leela unable to win games stockfish and alli were able. I mean stock and allie has more than 6 win games...
2838[10:57:09.984] ttomasagustin: @chetti95 why is it annoying to see better competitors?
2839[10:57:10.771] noobpwnftw: no chess engine can be perfect
2840[10:57:26.320] noobpwnftw: there are good days and bad days
2841[10:57:27.113] Cerenda: sure it can
2842[10:58:02.972] ttomasagustin: are we entering the danger zone?
2843[10:58:36.480] halberder: only the highway
2844[10:58:43.127] leotrubach2: is chat window too tall on you computer? I have to scroll down to write here and I can't see chat input and board at the same time.
2845[10:58:56.205] twindlers: tcecTB = perfect engine
2846[10:59:07.340] halberder: fine here
2847[10:59:45.525] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
2848[11:00:08.956] noobpwnftw: too tall here
2849[11:00:10.323] xoto10: See Options - Chat size at top of screen @leotrubach2
2850[11:00:12.840] twindlers: @leotrubach2 you can change the chat size via the options menu
2851[11:00:27.589] lito108: try to zoom out
2852[11:00:36.676] leotrubach2: thanks
2853[11:00:37.340] noobpwnftw: worked @xoto10 Kappa
2854[11:00:41.642] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2855[11:01:02.352] xoto10: SeemsGood
2856[11:01:21.119] ttomasagustin: highway to hell
2857[11:01:37.167] xoto10: i always used to have to make it smaller, now I make it bigger on the new site
2858[11:02:01.282] twindlers: it's kind of a silly UI option ... you have to scroll up, change the chat size, scroll back down to see if it's right, scroll back up and change it again, etc. Would be much easier to have the chat size slider right next to the chat window ... but oh well.
2859[11:03:05.669] Ipmanchess: website updated with Lc0 as new Nr1!
2860[11:03:13.259] halberder: there's a video of a fish vs a scorpion, "yellow puffer fish eats everything" on youtube
2861[11:03:44.184] halberder: as you might guess from the title, the fish wins
2862[11:03:56.056] twindlers: @Ipmanchess which website? And what's Nr1?
2863[11:04:09.534] Ipmanchess: !ipmanchess
2864[11:04:09.756] Nightbot: see http://www.ipmanchess.yolasite.com
2865[11:05:07.946] Ipmanchess: click on i9 7980XE
2866[11:05:56.932] twindlers: I don't see "Nr1" on that site?
2867[11:06:29.534] Ipmanchess: you need to scroll a little..
2868[11:07:46.636] twindlers: I put "nr1" in the search and it doesn't find it. Should I pick one of the other options under "i9 7980XE"? I just clicked it, but not one of the two options under it?
2869[11:09:20.584] gakuj505: PogChamp
2870[11:09:54.918] gakuj505: :) KappaPride KappaRoss Kappa
2871[11:10:40.828] KrzaQ2: Oh wow. I leave for a few days and Allie climbs to the top :O
2872[11:10:43.694] twindlers: I just googled "nr1" and it appears to be some kind of submarine
2873[11:10:43.903] Lolligerhans: SF still in? cmonBruh
2874[11:11:36.923] twindlers: SF about to go on a 33 minute and 15 second think here NotLikeThis
2875[11:11:41.944] xoto10: !bishop-pair scorpio sf
2876[11:11:42.145] Nightbot: scorpio is so doomed, sf has the bishop pair!
2877[11:12:12.258] xoto10: nr1 = number 1 = best
2878[11:12:13.168] alayant: "nr" = "number"
2879[11:12:38.858] twindlers: ah, thanks ... so Elsie is the new #1 submarine ;-P
2880[11:12:59.657] twindlers: I haven't finished my coffee yet :)
2881[11:14:54.524] twindlers: scorp's capture draws this game out another 10 plies ... trying to goad the fish into another crash!
2882[11:14:56.219] lrmead: @twindlers Who is "Elsie"?
2883[11:15:08.913] twindlers: say it out loud
2884[11:15:20.418] twindlers: "LC"
2885[11:15:30.321] twindlers: LC0
2886[11:15:49.403] kamboonseng: Why is BlueLeela so optimistic?
2887[11:15:49.966] ttomasagustin: i finally understand it
2888[11:16:25.388] twindlers: @kamboonseng blueleela incorporates crash odds into her PV ;-P
2889[11:16:58.535] ttomasagustin: scorpio wants to spice things
2890[11:17:17.275] kamboonseng: SF ain't gonna crashes...
2891[11:17:38.880] ttomasagustin: only time will tell
2892[11:17:43.963] twindlers: hope not ... but it definitely adds to the suspense of otherwise long boring dras
2893[11:17:50.642] twindlers: draws
2894[11:18:15.251] ttomasagustin: :D
2895[11:18:16.824] rocky640: @Ipmanchess is it possible to download reent games from your site ?
2896[11:18:20.850] kamboonseng: Do the programmers know what caused the crash?
2897[11:18:49.175] twindlers: they have a pretty good guess https://github.com/official-stockfish/Stockfish/issues/2229
2898[11:19:10.477] ttomasagustin: they believe it is some buggy .dll archive
2899[11:20:03.470] twindlers: hard to tell for sure until the dll is reverted to the older version and see what happens ... but apparently it's pretty reproducible
2900[11:22:16.768] trajkoskigoran02: Theare will be another crashes
2901[11:22:38.018] ttomasagustin: who knows
2902[11:22:38.340] twindlers: this post from @CoffeeOne 3 weeks ago is a bit ironic https://github.com/official-stockfish/Stockfish/issues/2229#issuecomment-524661162
2903[11:22:43.713] twindlers: "No news from my side, I requested support, but the STOCKFISH developers/maintainers are not interested."
2904[11:23:06.088] twindlers: I'm guessing the SF devs are interested now LUL
2905[11:23:19.546] Ipmanchess: @rocky640 are you on also on Lc0 discord
2906[11:23:44.317] rocky640: no
2907[11:24:53.126] Ipmanchess: Here a invite : https://discord.gg/xxrtU3
2908[11:25:46.026] Ipmanchess: a easy way to drop files also..when you enter find ipmanchess
2909[11:26:02.048] rocky640: ok
2910[11:28:24.718] Ipmanchess: i see a rocky online now..is that you
2911[11:29:41.541] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2912[11:30:29.077] rocky640: yes
2913[11:32:52.072] Ipmanchess: click left above on icon you will see you get message from me
2914[11:37:36.125] mariushav: SF has seen this game as dead equal from the very beginning
2915[11:39:51.346] kanchess: that neural eval needs a men in black deneuralizer
2916[11:39:52.800] noobs_killer: SF x=cant beat scorpion, end of era
2917[11:40:31.736] twindlers: someone should build a denueralizer network engine
2918[11:41:05.066] AfuroZamurai: only 18 and a bit games left for SF to survive
2919[11:42:02.480] twindlers: the NN evals make SF play many more moves on that faulty dll than otherwise would be necessary
2920[11:43:41.242] mariushav: What strength do you think would be plausible to achieve by training a pretty shallow / small net in the following way: 1) Show a FEN, make a direct prediction on the result of the game, and propagate the error back. Generate a few million FENS paired with results from a database of strong chess player's games. Would >2000 be feasible?
2921[11:44:08.880] Causa_Sui: Alliestein #1??
2922[11:44:24.882] sometimesithurts: !arsenal a sub type that cmnd for me pls
2923[11:44:30.798] mariushav: Really?! I thought it used RL TD methods?
2924[11:45:14.105] sometimesithurts: or is that a mod only cmnd?
2925[11:45:16.527] twindlers: @mariushav sounds feasible ... not sure how shallow/small of a net you could get. You might phrase it another way as "how shallow/small of a net could you get using that method and still have >2000 elo"
2926[11:46:34.798] sometimesithurts: @twindlers pls type !arsenal for me
2927[11:46:37.585] mariushav: @twindlers I'm not sure if >2000 would be plausible regardless net tho. No searching, no TD or otherways exploration of intermediate positions, just a cold prediction on 1 / 1/2 / 0
2928[11:46:49.843] twindlers: !arsenal
2929[11:46:50.124] Nightbot: This is @noobpwnftw 's arsenal: 4xE5-4669v4 88C/176T for BlueFish, 8xXeon Scalable 8168 192C/384T for Redfish, 100x E5-265(6)0 on FishTest (800 cores), 100x Ryzens on book building, 1400 cores for xiangqi book, and 20x ryzen 3700x for testing
2930[11:47:05.788] sometimesithurts: thx
2931[11:47:37.394] twindlers: @mariushav oh, I didn't realize no search was a requirement ... in that case you'd probably need a pretty large net
2932[11:48:45.524] twindlers: @mariushav you appear to already have your prime membership linked to your twitch account (based on the blue crown next to your name) ... you could get a free sub and have access to "sub only" commands yourself ...
2933[11:49:59.991] twindlers: @sometimesithurts yw
2934[11:50:04.088] snipertooth: danger time
2935[11:50:17.253] mariushav: @twindlers I'll look into that! :)
2936[11:51:15.620] NikolayIT: stockfish crashed?
2937[11:51:20.053] NikolayIT: this isn't possible
2938[11:51:34.272] twindlers: !prime
2939[11:51:48.567] kanchess: !define crash
2940[11:51:48.751] Nightbot: goto https://en.wiktionary.org/wiki/crash nadedKing
2941[11:52:13.170] twindlers: @kanchess why is !prime a mod-only command?
2942[11:52:28.377] twindlers: do you not want non-moderators to help you get subs?
2943[11:52:33.533] kanchess: reset it
2944[11:52:39.403] kanchess: try again
2945[11:52:44.974] twindlers: !prime
2946[11:52:45.203] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2947[11:52:50.791] kanchess: ty for noticing
2948[11:52:56.887] DanielShawul: i hope this position is not blunderable by scorpio
2949[11:53:13.128] Archstro: Go to TCEC they said... You're going to witness SF crashing they said...
2950[11:53:26.356] polkiujmn: @Archstro Hey it's gonna happen. Sonner or later.
2951[11:53:45.064] mariushav: @twindlers It appears to be linked to some other channel at the moment, but will make sure to switch over here once it's renewed!
2952[11:53:54.865] kanchess: yes you never know when surprised pop up
2953[11:54:14.472] chmariach: hang in there SF!
2954[11:54:36.746] Aloril42: Simple definition of engine crash: you can't continue game without restarting engine. Thus time loss is not crash. Illegal move would be border case, you might be able to continue by feeding same position again, just not with supplied move.
2955[11:55:18.641] tinduz_r: Drawish position, SF as block, kinda low in time, shuffling a bit, Scorpio not moving towards zero.... recipe for something?..
2956[11:55:22.065] ジブ940 (jb940): I would alter it slightly aloril
2957[11:55:46.749] ジブ940 (jb940): You can't continue the tournament without restarting or crashing an engine
2958[11:55:49.107] tinduz_r: okay.. scorpio says draw..
2959[11:56:02.252] kanchess: dr0.00w this 0.00lready SwiftRage
2960[11:56:13.666] ジブ940 (jb940): In case of an eternal hang, the game finishes normally, but in fact the engine has "crashed"
2961[11:56:22.517] polkiujmn: Here comes the crash Kappa
2962[11:56:27.328] chmariach: !-
2963[11:56:28.127] Nightbot: NEG-A-TIVE
2964[11:56:42.116] kanchess: after being held captive for so long as audience one starts longing for the blunder
2965[11:56:54.873] Aloril42: "If you can't continue game" -definition would cover also illegal moves but not time loss.
2966[11:57:00.839] kanchess: to make the waiting worthwhile
2967[11:57:06.797] twindlers: or you could say "crash, hang, or illegal move all count as one strike" and go back to the "3 strikes and you're out" rule
2968[11:57:30.366] ジブ940 (jb940): @Aloril42 but wouldn't cover an engine stuck in an infinite loop
2969[11:57:32.583] Aloril42: @ジブ940 Already covered with "Can't continue game", obviously you can't when engine is not responsive...
2970[11:57:56.960] ジブ940 (jb940): It just runs out of time. And it's arguable that running out of time is the end of that game.
2971[11:58:15.714] twindlers: @Aloril42 what if the engine goes into an infinite loop with <88ms left on the clock LUL
2972[11:58:32.295] chmariach: but you could also argue that a properly crashing engine just fails to move again, hence loses on time?
2973[11:58:34.427] kanchess: we could call these suddenly ending games also IPPONs Aloril42 borrowed from Judo
2974[11:58:38.811] ジブ940 (jb940): You'd have to define what the end of a game means to non programmers
2975[11:58:41.906] Aloril42: @ジブ940 Yeah, but running out of time and responding to "stop" with quick "bestmove foo" you can continue game.
2976[11:58:54.031] kanchess: Ippon Victory in one move, one point
2977[11:59:12.072] mariushav: SF deep in the tank. Hold your breath everybody!
2978[11:59:12.771] polkiujmn: Surely this move is the crash PogChamp
2979[11:59:16.467] DanielShawul: scorpio has to capture that pawn before blundering
2980[11:59:18.593] twindlers: @chmariach not really, cutechess detects the loss of signal in that case
2981[11:59:23.353] Aloril42: That is the crucial difference between time loss and hang.
2982[11:59:23.562] ジブ940 (jb940): @Aloril42 you're not wrong it's just your definition seems to run into the issue of what people understand of end of game.
2983[11:59:31.997] ジブ940 (jb940): So needs more text to explain
2984[11:59:46.112] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
2985[11:59:54.282] kanchess: !addcom !ippon Ippon = Victory in one move, one point
2986[11:59:54.753] Nightbot: @kanchess -> The command "!ippon" has been added successfully.
2987[12:00:12.233] Aloril42: Yeah, need explanation in any case to see difference between hang and time loss.
2988[12:00:38.603] ジブ940 (jb940): Yea, I guess the definition of engine still responds to commands works.
2989[12:00:42.719] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
2990[12:00:49.740] DanielShawul: the difference between hang & crash is as potato & potaeto
2991[12:00:50.504] Aloril42: If you look at log file, difference is about 1 line :)
2992[12:01:24.653] twindlers: I think most people understand what a "hang" is ...
2993[12:01:28.398] ジブ940 (jb940): Uhoh. 0-1
2994[12:01:41.586] alchemist888: @DanielShawul are you afraid of blunder from Scorpio here? This is an incredibly simple position.
2995[12:01:51.054] ジブ940 (jb940): @twindlers you'd be surprised how many people keep and kept arguing SF simply ran out of time.
2996[12:01:55.655] kanchess: https://en.wikipedia.org/wiki/Hang_(instrument)
2997[12:02:00.353] DanielShawul: well after the pawn catpure it is now
2998[12:02:34.022] alchemist888: even without pawns this is a draw
2999[12:02:43.879] twindlers: @ジブ940 I'm not surprised because I was in the chat when the arguments were being made. My point is that just using the word "crash" in the rule as it's currently written is vague ... so if the rule is "crash, hang, or illegal move" then that makes it pretty obvious
3000[12:03:13.759] kanchess: is there one term we can gther all of those under
3001[12:03:45.139] twindlers: @kanchess just define your own like "a 'strike' is a crash, hang, or illegal move ... 3 'strikes' and you're out"
3002[12:04:17.257] ジブ940 (jb940): What is this, baseball? Kappa
3003[12:04:26.836] CatoTheYounger_TCEC: it's not cricket
3004[12:04:29.458] twindlers: no, bowling ;-)
3005[12:04:30.879] chmariach: something like "failure to adhere to UCI protocol"?
3006[12:05:02.677] kanchess: "unruly behavior" :P
3007[12:05:07.359] tinduz_r: !ippon
3008[12:05:07.604] Nightbot: Ippon = Victory in one move, one point
3009[12:05:39.857] Aloril42: Hang: 23518997 >Stockfish 190826(3): stop Terminating process of engine Stockfish 190826(3)
3010[12:05:53.661] chmariach: "stockfish did a naughty"
3011[12:05:59.043] tinduz_r: Isn't "ippon" more fittingly used in judo on a single move that wins the game "very cleanly"? Winning by a crash you did not cause to your opponent seems unfitting for an "ippon".
3012[12:06:01.046] ジブ940 (jb940): We should also have rules for disrespectful or unsportsmanlike behavior. I csll it "shuffling", that should root out our pesky NN problem Kappa
3013[12:06:01.047] Aloril42: Time loss: 260671267 >RubiChess 1.5(143): stop 260674029 <RubiChess 1.5(143): bestmove f5g5 260674029 >RubiChess 1.5(143): quit
3014[12:06:12.731] Aloril42: Ok, so 2 lines, 1 more line.
3015[12:06:45.254] chmariach: unsportsenginelike
3016[12:07:19.554] chmariach: long think here
3017[12:07:22.146] K_recreant: crash
3018[12:07:35.073] AfuroZamurai: SF died again?
3019[12:07:43.753] kingscrusher: high level shuffling is like the celebration of non-commital decisions
3020[12:07:49.191] BehaardtheBonobo: we need a !sfcrash command with emotes
3021[12:07:53.857] tinduz_r: 3 minutes...
3022[12:07:57.511] snipertooth: scorpio killed SF
3023[12:07:57.660] RubiChess: @Aloril42 CPU usage?
3024[12:08:05.374] kingscrusher: neural networks though seem far more careful on type 1 commital decisions like a pawn break.. far more than AB engines as far as i can tell
3025[12:08:08.283] AfuroZamurai: okay, nvm it moved
3026[12:08:14.189] Aloril42: CPU at 100% (but you saw move already)
3027[12:08:17.016] paulcharlesmorphy: Will SF be able to qualify for thr SuFi?
3028[12:08:26.913] kingscrusher: so more extreme on type 1 preperation.. and celebrating type 2 almost indefintiely
3029[12:08:42.401] twindlers: if "strike" is too close to baseball then use another word like "fail", "a 'fail' is a crash, hang, or illegal move ... 3 'fails' and you're out"
3030[12:09:03.200] alchemist888: 3 fold soon
3031[12:09:14.280] marcobelli1979: what's happening to stockfish?
3032[12:09:21.756] paulcharlesmorphy: @kingscrusher I've just watched your analysis on Leela vs Stockfish in the French
3033[12:09:30.233] kingscrusher: cool
3034[12:09:45.475] octopoulo: pawns don't go back, right? :)
3035[12:09:53.165] paulcharlesmorphy: They are entertaining
3036[12:10:10.939] tinduz_r: Draw!
3037[12:10:12.451] alchemist888: Draw, SF avoided a crash
3038[12:10:23.418] Aloril42: !nocrash
3039[12:10:23.973] Nightbot: galacticAWW galacticDERP galacticDERP galacticDERP
3040[12:10:29.498] chmariach: !gg well played SF
3041[12:10:29.678] Nightbot: haaNT Good Game
3042[12:10:39.931] snipertooth: whew
3043[12:10:56.523] tinduz_r: Weird times, we congratulate SF for not crashing.
3044[12:10:58.372] Aloril42: (script) 93: 8. ScorpioNN 3716(+1) - 3. tcecSF 3831(-3) tcecDD https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-93-scorpionn-v3-0-1-n_maddex_int8-stockfish-190826/
3045[12:11:00.524] chmariach: that was tense
3046[12:11:06.990] alchemist888: Go Leela!
3047[12:11:08.117] ジブ940 (jb940): Can't we just add a 1% chance to crash for each engine each game? Look at the excitement it adds 4Head
3048[12:11:17.257] CCLS: leela needs to start winning
3049[12:11:25.618] octopoulo: The question is: "will SF remain in the top 3 chess engines in 2 years time"?
3050[12:11:26.495] ジブ940 (jb940): Sorry when I spoiled tcecs masterplan OpieOP
3051[12:11:37.593] noobs_killer: what is SF chance of survival ?
3052[12:11:43.979] alchemist888: @octopoulo easy
3053[12:11:53.373] Aloril42: (script) if 1-0: tcecLC 3843(+5), Kom: 3802(-5); if tcecDD : tcecLC 3836(-2), Kom: 3809(+2); if 0-1: tcecLC 3831(-7), Kom: 3814(+7); simulation and estimated start times: https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-94-speculation-simulation/
3054[12:12:02.148] fishtrawler: !reverse
3055[12:12:02.396] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
3056[12:12:05.610] eval_bot Game #66 "Komodo 2381.00" vs. "LCZero v0.22.0-nT40B.4-160": [book] 9. Na4 Re8 10. a3 Ba5 11. b4 Bb6 12. Nxb6 axb6 13. f3 Ba6 14. Bb2 Bxd3 15. cxd3 c5 16. Rc1 Qe7 17. Qc2 cxb4 18. axb4 c5 19. bxc5 bxc5 20. d4 cxd4 21. Bxd4 Qe6 22. Rfd1 h6 23. Rb1 Rec8 ... 1/2-1/2
3057[12:12:24.586] alchemist888: The only competitor for SF is Leela.
3058[12:12:40.216] noobs_killer: @alchemist888 . SF will be like Houdini in here in 2 years later
3059[12:13:01.573] tinduz_r: Just noticed Allie's Elo is higher than SF and Leela now.
3060[12:13:21.687] twindlers: The only competition to leels is Allie ;-P
3061[12:13:30.169] paulcharlesmorphy: Any news on Houdini 7?
3062[12:13:30.192] Aloril42: Decisive game here makes fairly big difference.
3063[12:13:40.355] ジブ940 (jb940): All haul our overlord Allie.
3064[12:13:47.999] ジブ940 (jb940): Hail. But haul works LUL
3065[12:13:54.191] alchemist888: A clone can't be a competitor :)
3066[12:13:55.255] ibn_akinfii: !revers
3067[12:13:58.427] octopoulo: @alchemist888 today, yes, but 2 years ago, Leela didn't even exist ...
3068[12:14:01.336] paulcharlesmorphy: !h7
3069[12:14:01.504] Nightbot: For h7 news when (if) it comes, such as this https://bit.ly/32n9WrN
3070[12:14:46.019] ジブ940 (jb940): Calling allie just a clone is like calling komodo an SF clone
3071[12:15:04.388] alchemist888: @octopoulo Do you think anyone would like to train an expensive neural network when there is already LCZero?
3072[12:15:42.265] alchemist888: Does Komodo use SF libraries?
3073[12:15:56.552] noobs_killer: !leela1-0
3074[12:15:57.004] Nightbot: A total of 349 wins have been cheered for Leela tcecK
3075[12:16:20.729] paulcharlesmorphy: Wait, so is Kibitzing Leela slower than playing Leela?
3076[12:16:30.986] ジブ940 (jb940): They al l use the same bit boards, the same zygysy code, the same threading libraries. I'm sure they implement atleast 50% of the same features.
3077[12:16:34.645] alchemist888: Komodo evals and seach is simular with SF?
3078[12:16:40.974] octopoulo: @alchemist888 good point
3079[12:16:41.021] Vizvezdenec: LLR: 0.08 (-2.94,2.94) [0.50,4.50] Total: 188942 Kappa Kappa
3080[12:17:14.559] gmfake: @ジブ940 allie and leela implement like 90% the same features lol
3081[12:17:15.221] ジブ940 (jb940): @alchemist888 right, and allie search and eval also has totally different code.
3082[12:17:27.270] CatoTheYounger_TCEC: @alchemist888 so no more new NNs on the horizon? they all gave up and ran away?
3083[12:17:35.879] alchemist888: totally different? :)
3084[12:17:40.532] khotan: ofc Allie is a clone
3085[12:17:41.274] ジブ940 (jb940): @gmfake you'd be surprised actually.
3086[12:17:42.866] frakty_zero: why is leela playing all the scotch games?
3087[12:17:46.222] Aloril42: I think weight loading code should be same. Less hassle for users.
3088[12:18:14.790] muppetmuppet: the openings are forced till the green ends
3089[12:18:19.892] paulcharlesmorphy: I love the Scotch
3090[12:18:44.445] CatoTheYounger_TCEC: Macallen?
3091[12:18:51.766] alchemist888: @CatoTheYounger_TCEC Train a network by generating games is one thing. Using games to train your network is different. The first is hundreds of times more expensive
3092[12:18:52.488] twindlers: @frakty_zero blame Cato
3093[12:19:13.909] frakty_zero: so far 3/4 of scotch games I ve seen were dead draws lmost right of the openning
3094[12:19:25.721] CatoTheYounger_TCEC: didn't answer my question
3095[12:19:55.600] muppetmuppet: let me look up humans and the scotch
3096[12:20:12.031] Aloril42: @CatoTheYounger_TCEC Winter is going to use NN next season.
3097[12:20:13.962] alchemist888: @CatoTheYounger_TCEC At least I haven’t heard anything about new interesting NNs.
3098[12:20:20.999] CatoTheYounger_TCEC: ok
3099[12:20:57.621] alchemist888: Allie is using LS net which using LCZero training process
3100[12:21:09.668] twindlers: wiNNter
3101[12:21:16.408] Aloril42: Hopefully Fire is hybrid next season, author already has strong AB engine so good opportunity.
3102[12:21:24.253] frakty_zero: this will be a dead draw soon as well
3103[12:21:27.513] muppetmuppet: 51% draws no worse than normal for humans
3104[12:21:40.535] tinduz_r: Would SF, Komodo, Houdini, etc. be classified as expert systems?
3105[12:21:47.141] CatoTheYounger_TCEC: 21% in Catobook...not promising anything, but...
3106[12:21:57.235] alchemist888: I expect that new strong neural networks will appear if there are new discoveries in the training process.
3107[12:22:14.943] twindlers: @tinduz_r nope ... they use heuristics
3108[12:22:17.886] CatoTheYounger_TCEC: exactly
3109[12:22:40.840] Aloril42: T60 from LCZero project, but next season might too early for that, maybe S18 DivP?
3110[12:23:01.100] CatoTheYounger_TCEC: next season will surely be too early for that
3111[12:23:22.340] tinduz_r: @twindlers Aren't heuristics hardcoded into by humans be classified as an expert system?
3112[12:23:26.424] muppetmuppet: I thought 1 year was expected before competitive
3113[12:23:37.371] CatoTheYounger_TCEC: sounds about right
3114[12:24:16.177] twindlers: @tinduz_r the way I learned the definition of "expert system" is that it's a rules based system ... not an AB search using heuristics ... so I wouldn't classify them as "expert systems"
3115[12:24:57.691] tinduz_r: AB is a different part though. It's the search. The expert system I think may apply more to its evals.
3116[12:25:11.655] Archstro: Leela project will soon come to a point where it's pointless to continue unless they get access to some seriously advanced hardware.
3117[12:25:33.727] flitzekacke: @Archstro There are still plenty of ideas to try out
3118[12:25:59.513] twindlers: @tinduz_r the evals a a result of heuristics, not a set of rules run through a rules engine
3119[12:26:09.445] marklefler: Go Komodo!
3120[12:26:09.455] d0rus: Current test is maxing out, but there are plenty of ideas to improve both training speed and max skill level of the NN. Easiest way would be a larger NN, but there are many other ideas too.
3121[12:26:09.773] Archstro: Assuming T80 will be even more complex than T60, it'll take 2-3 years to train with our amateur hardware.
3122[12:26:33.094] atanh: Damn. Snoozefest since Allie's last win, eh? Good thing I was snoozing then. TehePelo
3123[12:26:35.470] tinduz_r: @twindlers Like, if you code into the machine to give extrapoints for a fawn pawn or subtract when it's fortressy, you're setting rules known by a "chess expert".
3124[12:26:39.912] flitzekacke: There are other ideas than increasing netsize again. @Archstro
3125[12:26:40.350] d0rus: The fact they're still going with current test hugely surprises me.
3126[12:27:14.583] frakty_zero: @Archstro but in 2-3 years the available HW for deep learning should be much more powerful
3127[12:27:26.702] flitzekacke: @d0rus which test do you mean ?
3128[12:27:40.389] muppetmuppet: t 60 I assume
3129[12:27:51.950] tinduz_r: @twindlers The contrast is to machine learning approach, where you use data to get weights (or heuristics) for you.
3130[12:27:55.234] twindlers: @tinduz_r have you ever coded a rules-based engine (doesn't have to have been a chess engine ... any kind of rules-based engine would do)?
3131[12:27:58.976] atanh: !faketemp
3132[12:27:59.228] flitzekacke: T60 hasnt even had it's first LR drop yet. Hard to judge where its going, looking good so far though
3133[12:28:00.728] Nightbot: The current GPU temperature is 74 Kelvin KappaWealth
3134[12:28:22.162] fish_stock: this is a win for leela
3135[12:28:32.472] twindlers: @tinduz_r weights are not the same as human coded heuristics
3136[12:28:52.125] Archstro: AlphaZero's servers would train T60 to maturity in less than a week, I assume.
3137[12:29:01.121] flitzekacke: @twindlers SF eval isnt quite human coded entirely though. Lots of heuristics which were tuned using selfplay
3138[12:29:06.990] tinduz_r: @twindlers Yep, I agree. They're different that they are gotten from data computed using linear algebra.
3139[12:29:07.372] d0rus: @flitzekacke Test 40, althou i'm not even 100% sure they're still improving it with RL.
3140[12:29:12.588] CatoTheYounger_TCEC: +10 =7 -8 54% success rate, 28% draw rate
3141[12:29:27.036] fferen: Leela and the Lazy Lizard
3142[12:29:34.979] Fabichreportedich: @twindlers This is actually only kinda true. Most modern engines do some sort of automatic tuning, which then trains your evaluation heuristic via a logistic regression
3143[12:29:36.633] muppetmuppet: the actual lczero team have stopped t 40
3144[12:29:42.231] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
3145[12:30:04.829] CatoTheYounger_TCEC: +2 =1
3146[12:30:06.522] muppetmuppet: some people are continuing it to see what they can get jh for example
3147[12:30:10.680] paulcharlesmorphy: How long do we still have until a matured T60?
3148[12:30:26.420] flitzekacke: @paulcharlesmorphy Months :P
3149[12:30:26.641] twindlers: oh true, my mistake ... when @tinduz_r wrote "weights" it made me think of a NN weights file ...
3150[12:30:32.888] d0rus: @flitzekacke Yup looks like test 40 is still going.
3151[12:30:56.608] tinduz_r: @twindlers In one aspect though, they're weights and heuristics are the same that they just give "weight" to a "feature" you think is worth it.
3152[12:30:56.680] muppetmuppet: How long before leaving a NN in a dark room is illegal
3153[12:31:14.424] paulcharlesmorphy: @flitzekacke how many? :P 3? 9? 22?
3154[12:31:17.741] twindlers: @tinduz_r heuristics (whether human programmed or machine learned) are not the same as a rules based expert system
3155[12:31:19.451] d0rus: @muppetmuppet Really? I see new Test 40 nets released up to at least a few days ago.
3156[12:31:19.476] Aloril42: At S18 DivP time T60 training time is something like 9-11 months.
3157[12:31:22.451] CatoTheYounger_TCEC: already the law in Utah
3158[12:31:28.816] Fabichreportedich: @twindlers Which might be true. but looking at it, the difference between a NN as evaluation heuristic and the hand written one is only the feature selection for most modern engines, which is less than many people expect the difference to be
3159[12:31:33.427] flitzekacke: @paulcharlesmorphy I would say 2-3 tbh
3160[12:31:36.127] Vizvezdenec: http://talkchess.com/forum3/viewtopic.php?f=2&t=71822
3161[12:32:14.885] d0rus: @Fabichreportedich Leela is also the most successful MCTS engine.
3162[12:32:34.511] flitzekacke: @Fabichreportedich Somewhat true but Leela has many many more trainable parameters vs traditional engines. However, traditional engines have the speed advantage :P
3163[12:32:57.496] dtracers: Go leela keep booming
3164[12:33:25.637] kanchess: trolls talking chess online, such brave men
3165[12:33:29.780] pdsmike: Allie :O
3166[12:33:36.334] tinduz_r: @twindlers Wouldn't the use of heuristics as part of your rules still be classified under expert systems.
3167[12:33:57.712] twindlers: @tinduz_r show me the rules-based engine inside of SF
3168[12:34:14.078] twindlers: the code is open source, so you can send me a link
3169[12:34:43.662] polkiujmn: SF survives yet another game
3170[12:34:51.625] tinduz_r: A moment. Let me look it up.
3171[12:35:22.335] tinduz_r: !sf
3172[12:35:37.427] fferen: a5 gg 1-0
3173[12:35:46.941] stickboy77: Draw
3174[12:36:05.249] twindlers: @tinduz_r https://github.com/official-stockfish/Stockfish/
3175[12:36:30.660] kanchess: SuFi may turn out as AllieStein vs Stockfish for all we know atm
3176[12:36:57.868] stickboy77: The next decisive game may not be for a while
3177[12:37:00.896] twindlers: KMCTS will make a late surge
3178[12:37:33.874] nsousa254: !51
3179[12:37:34.077] Nightbot: In absence of further evidence, and as the logs didn't provide counter evidence, TCEC has no choice but to rule this as a tcecSF crash. See also https://github.com/official-stockfish/Stockfish/issues/2291 , https://github.com/official-stockfish/Stockfish/issues/2229 , and the Crash info tab - see also !74 and !dll
3180[12:37:41.078] nsousa254: !74
3181[12:37:41.301] Nightbot: Crash unless crash dump investigation says reason is not in tcecSF submission. See !51 and !dll
3182[12:37:49.394] nsousa254: !dll
3183[12:37:49.609] Nightbot: ""Should Stockfish team be allowed to replace libwinpthread-1.dll with non-buggy version?" AntonMihailov: The final results of the voting are now in. Out of 8 votes we have 6 "yes", 1 "abstain", and one "no". This final no is the veto. Voting is anonymous and TCEC team will respect that and the outcome of the authors' vote nadedKing
3184[12:38:38.454] tinduz_r: @twindlers Not going through the whole code but maybe enough to show an idea of it is something like around Line 300 of evaluate dot cpp. It says in comment "// Bonus if piece is on an outpost square or can reach one" and "// Penalty if the piece is far from the king"
3185[12:38:51.243] nsousa254: SF will make mincemeat of div1 next season LUL
3186[12:39:12.578] Tesseract_A: clonne?
3187[12:39:25.274] Vizvezdenec: yeah probably the only engine to stop leela clones onslaught Kappa
3188[12:40:42.890] kanchess: !define clone
3189[12:40:43.124] Nightbot: goto https://en.wiktionary.org/wiki/clone nadedKing
3190[12:42:01.281] atanh: !games
3191[12:42:02.703] eval_bot 93 played game(s): W:23(24.7%) / D:65(69.9%) / B:5(5.4%)
3192[12:42:14.210] tinduz_r: @twindlers Anyway, that's why I'm curious if it's proper enough to call SF an expert system.
3193[12:42:21.985] Archstro: There's a chance Allie can become the champion this season, then not be allowed to compete next season because of changed clone rules.
3194[12:42:48.393] Vizvezdenec: allie wouldn't be champion this season
3195[12:42:52.843] alchemist888: Stockfish like John Connor, exists to stop rise of the machines
3196[12:42:56.236] Vizvezdenec: stein completely sux vs usual leela net
3197[12:43:18.068] Vizvezdenec: it's only good to punish houdini/komodo because part of it learning was based on their games
3198[12:43:36.421] Vizvezdenec: basically it's like net trained to pass divP more or less while being weaker than usual
3199[12:43:36.852] d0rus: What engine is like allie that it's considered a clone?
3200[12:43:44.011] Archstro: I'm assuming Allie vs. SF as the most likely outcome of this season by the way things look right now.
3201[12:43:50.676] atanh: !cdb
3202[12:43:52.096] chessdbcn (16b) • Bg4 -43 ! (00-00) • <ChessDBCN>
3203[12:43:52.798] Vizvezdenec: it got to SuFi of 4c last season and got smashed by stockfish
3204[12:43:56.743] alchemist888: I think trade penalty helps LS a lot to beat weak engiens
3205[12:44:54.952] twindlers: @tinduz_r here's a good example of "Rules Based Expert System" https://www.cs.utexas.edu/users/novak/cs381k368.html vs. AB search https://www.cs.utexas.edu/users/novak/cs381k104.html
3206[12:44:57.129] kanchess: so, a new version can play in SuFi like SF has been doing for years
3207[12:45:29.657] gideon425: wow, SF survives two more games
3208[12:45:35.052] d0rus: SF making it to SuFI this season seems unlikely both because it has to defeat leela and no longer crash while doing so.
3209[12:45:39.128] kanchess: yo gideon425
3210[12:45:45.899] tinduz_r: @twindlers I'm saying though that they're not mutually exclusive. You can have an expert based system for the eval and have minimax AB for the search.
3211[12:46:01.015] gideon425: !gokan
3212[12:46:01.214] Nightbot: orbRIP
3213[12:46:01.222] kanchess: yes, maybe heading for AllieStein-Stockfish final :D
3214[12:46:09.947] kanchess: !godine
3215[12:46:10.115] Nightbot: slickentWindmill thickCHEF
3216[12:46:30.990] stickboy77: Allie Stein will not make the final unless SF crashes again
3217[12:46:37.080] twindlers: @tinduz_r ok, that makes sense
3218[12:46:53.488] tinduz_r: @twindlers In the same way that we have Leela have MCTS for the saerch and ML NN-based evals.
3219[12:47:15.936] d0rus: Idk about allie. It's expected to end lower but currently has a nice lead so it could be enough to sustain top 2.
3220[12:47:36.263] BehaardtheBonobo: @Vizvezdenec Don't forget that it has contempt which only works against weaker engines like K/H
3221[12:47:40.946] Archstro: SF will keep winning more games unless another crash happens. I'm not sure if Leela can keep up. She draws a lot.
3222[12:47:43.979] BehaardtheBonobo: Stein, I mean
3223[12:47:51.074] twindlers: @tinduz_r I don't think I'd classify SF's eval as a rules based expert system though ... but your definition may vary
3224[12:48:32.435] tinduz_r: @twindlers I'm kinda trying to form a case that it's not so much of a battle between AB vs NN but more of "expert system" vs "neural network". AB vs MCTS would be the proper battlefield for AB.
3225[12:49:43.722] tinduz_r: @twindlers Allie represents NN + Minimax AB, that's why it would be too muddled to keep it AB vs NN for me.
3226[12:49:47.141] twindlers: @tinduz_r oh yeah, the whole AB vs. NN thing is a bit misguided ... AB is for search whereas NN is for eval.
3227[12:50:14.270] twindlers: I think Giraffe was an AB engine with NN eval
3228[12:51:12.473] twindlers: really, Allie uses AB? someone in here was just calling Allie a clone of Leela
3229[12:51:41.258] atanh: Technically, A/B engines are a product of NN training as well: biological NNs. TPFufun
3230[12:51:43.378] tinduz_r: The stein net has a lot from Leela, plus other parts of the NN code.
3231[12:52:10.018] kanchess: gideon425 i have renamed your go to a more intellegible command
3232[12:52:40.899] gideon425: !gogod
3233[12:52:46.270] kanchess: almost :O
3234[12:52:55.375] gideon425: !gogodlike
3235[12:53:02.885] kanchess: !gogid
3236[12:53:03.133] Nightbot: slickentWindmill thickCHEF
3237[12:53:44.068] gideon425: that is more stylish
3238[12:53:49.540] kanchess: thought so too
3239[12:54:22.648] alchemist888: @twindlers Can't the clones use something of their own? That's why they are clones, and not a complete copy. Did Allie + Stein net beat LCZero v0.22 32930 to show how extra AB minimax works well?
3240[12:54:25.624] gideon425: and anshul appreciated his command the other day
3241[12:54:34.136] kanchess: !goanshul
3242[12:54:34.312] Nightbot: lettalHip
3243[12:54:39.858] kanchess: nice to hear :)
3244[12:54:48.066] mdt_sf: !sfprob
3245[12:54:48.265] Nightbot: see https://i.imgur.com/HUYlRze.png by lmabacus
3246[12:55:19.944] twindlers: @alchemist888 which part of Allie is a clone of Leela?
3247[12:55:30.525] glbchess64: According to @jjosh (lco discord) : stein net is TCEC + CCRL games + lot of T30 games
3248[12:55:41.332] alchemist888: @twindlers see talkchess theme
3249[12:55:54.631] noobs_killer: !1-0 , leela likely get ticket for final
3250[12:55:58.818] Twipply: @twindlers I'm not 100% sure, but I think Giraffe actually used its NN for move ordering
3251[12:56:06.959] Twipply: Definitely rather you look it up than trust my word though
3252[12:56:21.227] twindlers: but if Allie uses AB and Leela uses MCTS ... and the Stein net is different from Leela's net ... then that would hardly be a "clone" wouldn't it?
3253[12:56:24.252] gideon425: NNs are good for leaf evaluation
3254[12:56:40.896] The_Boomer123: Who created Alliestein? It has gotten really good really fast
3255[12:56:54.985] alchemist888: @twindlers Allie uses MCTS + some AB minimax, but main is MCTS
3256[12:57:01.920] mdt_sf: eval is always applied to leaf nodes
3257[12:57:13.407] Twipply: Seems the wiki says its NN is for eval
3258[12:57:32.960] mdt_sf: i meant to say, eval is only applied to leaf nodes
3259[12:57:39.778] The_Boomer123: Allie vs Leela in the SUFI would be very interesting
3260[12:57:56.245] kanchess: as would Allie vs SF be
3261[12:57:59.478] tinduz_r: Leela might not even make the sufi.
3262[12:58:04.311] kanchess: something different
3263[12:58:07.067] Twipply: I could've sworn I read it did move ordering instead/as well
3264[12:58:09.095] kanchess: to shake things up
3265[12:58:43.775] twindlers: @Twipply doesn't eval affect move ordering?
3266[12:58:56.415] d0rus: Allie, leela or Sf might not make it to SuFi!!!!
3267[12:59:25.416] kanchess: Stoof vs Scorpio sufi you mean
3268[12:59:27.057] d0rus: Big surprise, but that's a fact actually.
3269[12:59:44.532] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
3270[13:00:02.382] tinduz_r: Yep. Weird tension we have on the Top 3.
3271[13:00:04.498] twindlers: KMCTS is going to crash twice vs. K and put K into SuFi tcecK
3272[13:00:14.755] kanchess: :D
3273[13:00:20.187] Path2Grandmaster: IMHO, This is the best TCEC and the best CCC yet!
3274[13:00:42.646] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
3275[13:01:57.428] fferen: dead draw ResidentSleeper
3276[13:02:46.505] kanchess: !bishop-pair each everyone
3277[13:02:46.684] Nightbot: each is so doomed, everyone has the bishop pair!
3278[13:04:51.474] Stephane_Nicolet: Go Komodo!
3279[13:06:57.188] frakty_zero: so are the opennings selected randomly for each unique engine pairing? Or how does it work? Was it just bad luck that leela got 2 scotch games x reverse?
3280[13:10:03.251] alchemist888: @frakty_zero Cato fault
3281[13:10:46.644] Twipply: I just skimmed the Giraffe paper
3282[13:10:53.711] Twipply: Turns out they did ML stuff for both eval and move ordering
3283[13:10:56.934] Twipply: so I guess we were both right
3284[13:11:39.448] frakty_zero: @alchemist888 so is it random?
3285[13:12:03.037] kanchess: !randomowl random it is
3286[13:12:04.017] Nightbot: athFA
3287[13:12:05.969] alchemist888: @frakty_zero should be - yes, otherwise it would be scam :)
3288[13:12:21.496] fferen: !blamecato
3289[13:13:19.491] tinduz_r: !catosfault
3290[13:13:19.683] Nightbot: someday, the eulogy will go, "it's HIS fault that he died--he's to blame! he was always to blame! he led a blameful life! to his credit, he never denied it, though it is his fault that he has left us without admitting all the scams he masterminded" --CatoTheYounger_TCEC
3291[13:13:23.491] marklefler: if anyone has some clever numeric software, I was wondering how often the game is a raw when Komodo says it is? How accurate is its draw score? Does it generally see draws sooner than other programs?
3292[13:14:28.405] gobbledydook: what's with stockfish this season?
3293[13:14:31.072] tinduz_r: Redmodo in the past seem to call draws better, but maybe I'm misremembering.
3294[13:14:32.383] Twipply: I imagine such questions would fall under the issue of "Which engine is right?"
3295[13:14:43.910] Twipply: If Komodo says 0.00 and SF says 1.00 at some point, and it ends up being a draw, was Komodo right?
3296[13:14:48.669] Twipply: Or was SF right and it just blundered?
3297[13:15:09.107] gobbledydook: so many crashes
3298[13:15:11.642] noobs_killer: b3, Nb2, Nc4 will be a good knight relocate
3299[13:17:06.086] alchemist888: Hmm, good question. I would say that Komodo sees draws more often than other programs. It's like: draw, draw, draw... woops it's not a draw anymore :)
3300[13:17:13.412] Y_h_w_h: @marklefler if you search for statistics software then use R
3301[13:18:59.925] atanh: Damn, lookin' drawish. :(
3302[13:19:35.692] tinduz_r: The more luck-based parts that exist in a competition, the less value it's result has in terms of merit. There's at least two I'm seeing right now, who gets more decisive openings and who SF crashed against and gave its opponent a free point. Maybe there would be ways to reduce luck-based parts.
3303[13:19:48.943] alchemist888: It seems to me the question of how accurately the engine finds a draw or win directly depends on its strength. And individually, these questions are meaningless.
3304[13:20:27.701] gonzochess75: WOW, Allie in quite a hot streak!!
3305[13:21:33.932] gonzochess75: Allie has highest live elo!!
3306[13:21:37.438] Ove__: !hypealliestein
3307[13:21:37.640] Nightbot: fl0mHype bueRocket VoHiYo fl0mHype
3308[13:23:06.439] Twipply: Looks like Allie and SF should be tied 15 points each leaving poor LCZ out of the superfinal
3309[13:23:08.977] Twipply: Cool stuff
3310[13:23:16.959] Tesseract_A: lmao my homeSF crashed
3311[13:23:31.565] tinduz_r: Don't jinx it. :P
3312[13:23:44.062] CatoTheYounger_TCEC: poor LCZ
3313[13:23:58.718] Twipply: Poor LCZ indeed
3314[13:24:08.884] Twipply: I was 'worried' about this after it didn't qualify for the CCC final
3315[13:24:09.824] Tesseract_A: SF gui was analyzing a position from a leela game and leela found an amazing bishop sac, it took SF a while to see it but Bg5 worked in it
3316[13:24:11.275] Sounorro: so her time has come?
3317[13:24:13.348] alchemist888: It's your fault, Cato :)
3318[13:25:00.745] Twipply: Though ever since Komodo shifted effort onto the MCTS version I've been wondering if it was putting itself out of the superfinal
3319[13:25:13.808] Twipply: ahead of LCZ
3320[13:27:09.209] dtracers: come on leela break through
3321[13:27:20.035] Tesseract_A: anyone know what this line is about; it crashed my homeSF ''Crashing on exception: *** -[__NSArrayM insertObject:atIndex:]: object cannot be nil''
3322[13:27:52.623] Tesseract_A: Termination Signal: Illegal instruction: 4
3323[13:28:21.118] Stephane_Nicolet: @Tesseract_A it is a Mac thing
3324[13:28:45.978] Stephane_Nicolet: what do you mean exactly by "my homeSF"?
3325[13:28:58.123] Tesseract_A: it's SF in a GUI
3326[13:29:13.189] gonzochess75: That GUI is a Mac GUI right?
3327[13:29:15.580] marklefler: We have actually wokred more on regular K recently, and things that benefit both search types. What happenes each day just depends on what we think about. For example, about a week ago Larry noticed we were something like 20 elo stronger in fixed movetime game against Stockfish than in base + increment, suggesting we were missing elo with the current time allocation scheme.
3328[13:29:18.611] gonzochess75: cuz that is what is crashing
3329[13:29:28.048] Tesseract_A: SF10 mac app yes
3330[13:29:42.494] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
3331[13:29:46.998] gonzochess75: There you go
3332[13:29:49.623] Stephane_Nicolet: this GUI: https://stockfishchess.org/mac/ ?
3333[13:30:00.900] Tesseract_A: indeed
3334[13:30:22.565] gonzochess75: @marklefler looks like regular K might be your guys best ride back to Sufi
3335[13:30:30.828] Tesseract_A: if i send the error report what comment should i leave if anything? i had it running for about 14 hours
3336[13:30:31.922] Twipply: @marklefler Cool. Glad to see regular Komodo is still getting some love
3337[13:30:32.238] Stephane_Nicolet: then it means there is a bug in the GUI, which is programmed in ObjectiveC
3338[13:30:42.459] marklefler: The Stockfosh team could go back to an earlier crash free Stockfish version, then stop development, because changing anythong could introduce crashes! :-)
3339[13:31:02.926] Stephane_Nicolet: it is not a crash in the engine, we don't use NSArray in the engine
3340[13:31:10.212] Tesseract_A: stop development and wait for libraries to change Kappa
3341[13:31:19.760] Twipply: I'm glad that K MCTS exists, it's nice to see engines of many varieties in my opinion
3342[13:31:22.376] gonzochess75: We are all just happy to see regular K getting stronger
3343[13:31:34.186] LoIiSquad: Yeah, KMCTS is cool
3344[13:31:45.019] Tesseract_A: @Stephane_Nicolet ok, then if i send the error report, will it go to SF people or mac people?
3345[13:31:49.103] marklefler: @Tesseract_A that is another solution. The key is stopping development! :-)
3346[13:31:58.369] gonzochess75: Hehe
3347[13:32:00.606] tinduz_r: Haha. Good ol' ribbin' in the competition by authors.
3348[13:32:30.092] gonzochess75: I am sure @marklefler has nothing but the good of SF in mind ;)
3349[13:32:34.050] Twipply: On a related note, I hope all go engine development doesn't shift to NN
3350[13:32:39.463] Twipply: I want traditional engines to live on, too
3351[13:32:48.658] Twipply: Likewise for chess if the NNs ever do fully take over
3352[13:32:56.750] alexey305472: K MCTS will be degraded in league?
3353[13:33:09.186] Stephane_Nicolet: @Tesseract_A Daylen Lee is the author of that GUI, if I remember correctly
3354[13:33:21.568] CatoTheYounger_TCEC: Twipply drives a Ford Model T
3355[13:33:25.546] Twipply: @alexey305472 bottom two engines get relegated
3356[13:33:28.688] gonzochess75: Largely depends on whether SF crashes again
3357[13:33:43.332] gonzochess75: if it does it automatically takes one of the relegation spots
3358[13:33:54.398] Stephane_Nicolet: Put a crash report there in the "issue" tab: https://github.com/daylen/stockfish-mac/wiki/User-Guide
3359[13:33:55.589] Tesseract_A: @Stephane_Nicolet so it should go to him? :P
3360[13:34:19.317] tinduz_r: @gonzochess75 Question, how much of Allie is MCTS vs AB now?
3361[13:34:20.011] Stephane_Nicolet: ^
3362[13:34:35.833] alexey305472: yes, K MCTS and Scorpion are definitely much weaker, and getting check mated every now and then. SF got checkmated only once. By Leela.
3363[13:34:37.210] gonzochess75: It does both
3364[13:35:03.224] tinduz_r: @gonzochess75 In terms of percentage of the time, do you have some rough numbers?
3365[13:35:16.500] noobs_killer: Draw against Komodo is a failure for Leela.
3366[13:35:18.788] gonzochess75: hard to put a percentage on it. Probably 85% mcts
3367[13:35:29.129] gonzochess75: 15% ab
3368[13:35:35.771] tinduz_r: Thanks!
3369[13:35:43.131] Twipply: @marklefler have you been looking up prominent Go engines or their authors for MCTS ideas/advice?
3370[13:35:46.569] alexey305472: Stockfish, even if it gets degraded, it will wreak havoc in weaker league. Weaker engines gonna suffer badly vs the Fish!
3371[13:35:59.536] Twipply: Presumably they'd know more than chess engine authors
3372[13:36:09.327] jjoshua2: scorpio also does some minimax
3373[13:36:20.435] lado933: Poor allie if SF makes it to sufi
3374[13:36:25.580] VetustaMorla1970: 6 asolated pawns!
3375[13:36:41.495] gonzochess75: @lado933 don’t feel bad for us!!
3376[13:36:41.784] jjoshua2: has @marklefler if K MCTS does any minimax of nodes that have enough MCTS visits?
3377[13:37:12.358] jjoshua2: i dont think we have any details of that level of stoof either
3378[13:37:37.038] gideon425: next season Allie moves to hand-written net
3379[13:38:01.511] tinduz_r: "hand written"?
3380[13:38:04.045] Stephane_Nicolet: @jjoshua2 That's state of the art for strong Monte-Carlo programs?
3381[13:38:11.693] atanh: !deciboom
3382[13:38:12.031] Nightbot: ?ʙᴏᴏᴍ
3383[13:38:41.165] alexey305472: I don't believe it's possible to hand write a Neural network. it goes against the teaching of Google Alpha Zero, and I doubt results will be good.
3384[13:38:50.155] GamBito93: @Stephane_Nicolet: could you please make more ''eval'' testing on Fishtest now (not only search changes) in order to improve Stockfish evaluations? With 600 machines now, much more tests can be done. BTW, Thanks for your great work on Stockfish!
3385[13:38:58.218] fferen: !femtoboom
3386[13:39:22.326] tinduz_r: A net need not be made using Deep Learning by Google always.
3387[13:39:28.011] Twipply: @alexey305472 big network? No. Tiny little simple network? Sure
3388[13:39:56.291] dtracers: Go leela!
3389[13:40:01.024] Twipply: I imagine one of the first networks anyone starting to learn will see is that of a XOR
3390[13:40:05.197] Twipply: You can write that by hand trivially
3391[13:40:08.766] gonzochess75: Stein net is already very different. SL vs RL
3392[13:40:17.057] NikolayKralev: does SF still has chances ?
3393[13:40:32.266] gonzochess75: Stein net is not done the same way as A0
3394[13:40:50.641] tinduz_r: I'm actually kinda happy people are exploring non-zero versions while Leela team is exploring how far the "zero" concept can go.
3395[13:41:06.836] gonzochess75: @NikolayKralev if it does not crash again, sure
3396[13:41:11.187] Twipply: I don't much like the zero concept at all
3397[13:41:19.945] gonzochess75: i would even put it as favorite for Sufi if it does not crash again
3398[13:41:33.599] Twipply: and as far as I can tell, the DeepMind paper was lying/misleading about Zero being stronger
3399[13:41:47.164] JISICHANCZ: Where is Gonzo chess ??
3400[13:41:50.296] tinduz_r: It's still a worthwhile experiment to see though. How far machines can self-learn.
3401[13:41:53.900] gonzochess75: @Twipply me either. And to be fair creators of leela say it already is non zero
3402[13:42:08.373] Stephane_Nicolet: @GamBito93 in fact we have been testing like crazy in fishtest during the last two weeks, and indeed there has been an implicit surge among the devs to switch to more eval patchs (as opposed to search patches or simplification patches) -- I am very pleased of this, but I can't give orders to my collegues, you know
3403[13:42:16.447] muppetmuppet: Skynet is where it ends obviously
3404[13:42:19.508] Twipply: I'd even flat out reject that zero is stronger
3405[13:42:21.420] tinduz_r: Leela is non-zero already?
3406[13:42:40.661] Archstro: Zero concept requires an obscene amount of computing power.
3407[13:42:44.703] gonzochess75: Sure. It uses tablebases for jnstnade
3408[13:42:48.672] georgelvoinicel: SF poor performance in French against Lc0 seems to be set in stone.
3409[13:43:03.679] muppetmuppet: tblebases are just part of chess
3410[13:43:03.930] Twipply: I'd easily put EGTB in the zero category
3411[13:43:32.010] JISICHANCZ: @gonzochess75 Will Allie win TCEC 16 ??
3412[13:43:36.077] gonzochess75: Zero means no chess specific domain knowledge
3413[13:43:38.162] muppetmuppet: it probably can't pronounce their stupid words
3414[13:43:38.986] tinduz_r: Tablebase scoring still seems in line with zero.
3415[13:43:52.476] gonzochess75: @JISICHANCZ keep watching and find out!!
3416[13:43:58.685] jjoshua2: @Stephane_Nicolet i think state of the art is using lower confidence bound mixed with MCTS
3417[13:44:13.406] jjoshua2: with possibly some regret minimilzation
3418[13:44:20.713] jjoshua2: leela go has experimented with a lot of these ideas with some gain
3419[13:44:33.653] muppetmuppet: I thought zero just meant no human interference with knowledge
3420[13:44:41.920] tinduz_r: At zero, you taught it the rules of what's a win and what's a loss. TBs just extended that.
3421[13:44:42.856] jjoshua2: leela go actually merged some lower confidence bound stuff into master
3422[13:44:59.992] JISICHANCZ: @gonzochess75 How many ELO Allie gained from the season 15 ??
3423[13:45:19.030] glbchess64: Zero concept is necessary for chess because human are too weak at chess. But you can have mixed approach based on good games from previous engine (SL) and games then produced with RL.
3424[13:45:26.496] gonzochess75: Dunno. But you can see live elo just like rest of us :)
3425[13:45:47.368] Twipply: Zero is necessary for chess and yet SF exists?
3426[13:45:52.186] glbchess64: It is the way Stein net is trained.
3427[13:46:01.032] gonzochess75: She is food. Is she SUFI champion good? Have to wait and see
3428[13:46:10.572] gonzochess75: She is good. She is not food.
3429[13:46:19.311] gideon425: SF will eat her up
3430[13:46:31.343] tinduz_r: Family friendly chat. :P
3431[13:46:34.105] gonzochess75: Very possible
3432[13:46:39.669] glbchess64: Necessary for NN chess
3433[13:47:02.790] Twipply: Why necessary for chess and not go?
3434[13:47:03.539] muppetmuppet: sf is actually not winning even without crashes
3435[13:47:03.784] gideon425: !addcom !food "Allie is food - gonzo"
3436[13:47:14.746] gonzochess75: Lol
3437[13:47:44.669] noobs_killer: !Ali SF era ended, check Ali art..
3438[13:47:44.865] Nightbot: All Artworks for TCEC https://www.flickr.com/gp/181644265@N06/563wXp
3439[13:47:50.716] Archstro: I don't like the idea of new ideas for Leela needing a year to be tested because of how long zero approach training takes with current hardware power.
3440[13:48:06.349] tinduz_r: You authors ought to be naming your engines creatively so we get more Rodent vs Cheese matchups.
3441[13:48:22.367] Twipply: Cheese vs Stoofvlees
3442[13:48:34.020] muppetmuppet: rock paper scissors lizard spock are good names
3443[13:48:35.358] glbchess64: Here we speak chess, not go. But if you want you can generalize to what ever game too complicated to human.
3444[13:48:49.989] Stephane_Nicolet: @jjoshua2 tahnks, I will have a look
3445[13:48:54.406] NikhilChess: !AllieStein
3446[13:48:54.683] Nightbot: See !allie and !leelenstein
3447[13:49:08.632] jjoshua2: @Stephane_Nicolet what elo was the mcts fork of SF?
3448[13:49:24.873] NikhilChess: !allie
3449[13:49:25.027] Nightbot: Allie is a neural network based engine written by Adam Treat, inspired by the AlphaZero paper and the Lc0 project. See - https://github.com/manyoso/allie , https://www.patreon.com/gonzochess75 and http://www.chessdom.com/alliestein-the-new-neural-network-entering-tcec-s15/ , Join the discord channel to follow Allie - https://discord.gg/UnbsJRU
3450[13:49:25.782] jjoshua2: you had a very promising looking fork that languished a while i guess
3451[13:49:38.951] jjoshua2: maybe time to bring it back with a few new ideas :)
3452[13:49:46.337] Sounorro: if any of my chess mentors saw me playing that structure of pawns for the black I'd certainly receive a kick in the balls
3453[13:50:18.208] muppetmuppet: you missed the game before with black moving all pawns to rank 6 except one
3454[13:50:24.041] Nuclearball: But your balls would be so tough you'd break their legs LUL
3455[13:50:32.067] Stephane_Nicolet: https://pdfs.semanticscholar.org/8119/ce3360001f46c4b0a95a98e9121ffcd89dde.pdf
3456[13:51:02.411] jjoshua2: i saw bernstein sampling mentioned but it looked complicated
3457[13:51:10.332] Stephane_Nicolet: @jjoshua2 abysmal, it is still a prototype
3458[13:51:12.213] jjoshua2: and i dont think anyone has gotten it to be strong
3459[13:51:19.733] tinduz_r: @muppetmuppet Yep, and SF evaluating it at negative out of book.
3460[13:51:51.643] jjoshua2: there also was an infinte armed bandit
3461[13:51:57.330] fire2fire: what wasted LC0 community resources if Allies is give better result
3462[13:52:04.818] jjoshua2: https://arxiv.org/abs/1803.04665
3463[13:52:18.812] totally_disconnected: Am I here?
3464[13:52:35.174] jjoshua2: what to do if an infinitely armed bandit is after you LUL
3465[13:52:52.408] DanceWithMeQt: dont they use NN from LC0 or something?
3466[13:52:54.600] totally_disconnected: Tell him to go climb a tree(3)
3467[13:53:00.313] muppetmuppet: give it underarm deodorant
3468[13:53:01.145] jjoshua2: https://github.com/leela-zero/leela-zero/issues/860#issuecomment-519879849
3469[13:53:30.530] jjoshua2: does a chess monster have enough arms to call it infinite?
3470[13:53:50.524] jjoshua2: @Stephane_Nicolet this github link is important for you tos see then
3471[13:53:53.186] alexey305472: how Allie does training? I mean it takes about 50x RTX 2070 to train Leela for months to get the results we're getting. How much GPU power is behind Allie?
3472[13:54:01.458] muppetmuppet: not sure anything in nature has more than 2
3473[13:54:07.271] jjoshua2: just a single 2080 ti @alexey305472
3474[13:54:31.337] jjoshua2: but it relies on all the games from ccrl and tcec and otehr contributors like sergio and leela t30 too
3475[13:54:49.492] alexey305472: how can Allie get away with single GPU and achieve such great results?
3476[13:54:55.417] jjoshua2: there is also a lot of postprocessing done using a threadripper and a bunch of SSDs
3477[13:55:03.131] jjoshua2: but its all one high end workstation
3478[13:55:21.801] jjoshua2: althoguh the generation of the public datasets was an immense work of the public computing
3479[13:55:32.512] jjoshua2: just ccrl is something like 1000 CPU years at least im sure
3480[13:55:33.597] LeonardoDiCaprino: is SF still stronger on an infinite budget?
3481[13:55:56.142] tinduz_r: Is the use of T30 and not T40 a choice based on something?
3482[13:55:58.648] lrmead: C'mon Leela you need this win ...
3483[13:56:07.993] atanh: @gonzochess75 So they're finally admitting TBs are non-zero, eh? I argued against TB rescoring from the start because it's non-zero. So many tried to argue it's zero. Yet, they could have easily developed a non-zero solution to it: Have part of Leela's training be to develop her own EGTB strictly from her own self-learned knowledge of the game. Wouldn't listen. Now they have to admit it. FailFish
3484[13:56:20.588] jjoshua2: i put T30 data in before i new how to get policy data from ccrl @tinduz_r
3485[13:56:46.350] Twipply: @atanh why do you think EGTB aren't zero?
3486[13:56:50.521] talkshowwcharles: Anything could be better with inf budget.
3487[13:56:51.821] alexey305472: I have a bunch of strong servers to donate time to Allie and other engines, possibly BrainFish. 112-CPU monsters, Dual Xeon 8176 + Dual GeForce RTX 2060 each server. my email is al4321@gmail.com
3488[13:56:51.851] jjoshua2: you can see dkappe train nets with any t30 data with good plicy data as well as dark queen
3489[13:57:17.428] jjoshua2: really?
3490[13:57:30.707] jjoshua2: how do you get all these computers
3491[13:57:54.261] muppetmuppet: he means wld sf be stronger than the other engines if both had infinite money
3492[13:58:03.543] LeonardoDiCaprino: ^
3493[13:58:04.384] Lord_Aldrich_: you guys talking about (computer) chess here? i thought it wasn't allowed.
3494[13:58:10.718] jjoshua2: there is actually some tuning verification that allie needs
3495[13:58:21.280] lrmead: @muppetmuppet Who has infinite money?
3496[13:58:29.920] jjoshua2: i think i have some new settings for gonzo to use in allie but he doesnt have the computer power to verify them i guess
3497[13:58:34.323] muppetmuppet: I think sf might be the strongest then its a thought exerceizse
3498[13:58:40.881] ka2ma2: wth happened to the leaderboards?
3499[13:58:41.058] Nightbot: What the Houdini erckMovement
3500[13:58:52.962] jjoshua2: lc0 would be the strongest assuredly
3501[13:59:09.526] jjoshua2: you could just train a whole new net from each position if you were a billionare playing CC
3502[13:59:10.793] tinduz_r: TPUs all over.
3503[13:59:15.802] muppetmuppet: as far as I know leela struggles to use many GPU so it is restricted to the best say 4 of those whereas sf is pretty unrestricted ?
3504[13:59:45.987] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
3505[13:59:46.692] muppetmuppet: oh I see
3506[13:59:48.643] ka2ma2: lol, there is a high chance we will get first 2x nn final
3507[13:59:49.551] Fabichreportedich: If you have infinite money you would have infinite computing power which means every engine can see until the end of the game tree, so they would all be equally good
3508[13:59:55.946] jjoshua2: if you are training a new net with the starting position of the current position of hte game there si no limit on the compute you can use
3509[14:00:05.938] jjoshua2: it also will be very good at detecting fortresses
3510[14:00:09.492] alexey305472: About Brain fish or brain Allie or brainLLeela : I do think it can improve chess performance significantly at the opening and early mid game, much better than human made opening books.
3511[14:00:11.686] talkshowwcharles: Leela could be fixed for more gpus if they hade strong incentive
3512[14:00:44.826] Sounorro: @Fabichreportedich that's technically unfeasable. there are more legal chess positions than atoms in the known universe.
3513[14:00:44.925] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
3514[14:01:52.407] jjoshua2: @Sounorro thats more of an argument that there does not exist infinite resources than that infinite resouces couldnt solve it
3515[14:02:10.193] gonzochess75: Zero push for lc0 against K here
3516[14:02:17.280] gonzochess75: !next
3517[14:02:17.466] Nightbot: Next DivP CPU engines + top 4 CPU engines from League 1 Bonus (Laser, tcecLC CPU, Ethereal and Xiphos), 30min+5s, 1x DRR, a bit over 2 days. Deadline for Sufi is last move of Bonus. Then testing for Sufi. Then Sufi.
3518[14:02:36.639] muppetmuppet: I suspect he wants to know if you have say $1 million for the computing parts which wld be better
3519[14:02:37.024] Twipply: @jjoshua2 presumably you could train a whole new Stockfish for each opening, too
3520[14:03:04.545] jjoshua2: with infinite resouces yes, but not with just a million dolars
3521[14:03:10.422] muppetmuppet: the question is which scales better I think
3522[14:03:43.975] muppetmuppet: he seems to have gone quiet though
3523[14:04:06.053] Sounorro: @jjoshua2 he spoke about money. money is nowadays a number inside a computer
3524[14:04:12.616] alexey305472: forget about 'infinity' as it would result in a 32-man table base and solved chess. let's speak about reality. Real hardware, say dozens of strong servers.
3525[14:04:53.636] gonzochess75: For how long?
3526[14:04:55.024] jjoshua2: lc0 should work better to combine multiple servers since NPS is 1000x lower the combination problem is 1000x easier
3527[14:05:06.826] tinduz_r: Dozens of TPUs could train a zero engine to unimaginable heights.
3528[14:05:13.413] gonzochess75: anton would love stronger machines for TCEC I bet
3529[14:05:19.413] jjoshua2: but both lc0 and SF have almost zero development work torwards clusters, SF has more than zero so it would win currently for that reason only
3530[14:05:53.162] muppetmuppet: oh yes I forgot TPU's
3531[14:05:57.773] jjoshua2: however the lc0 training would be reusable at CC level
3532[14:06:10.730] jjoshua2: you can even just fine tune a net only with low learning rates for the current position
3533[14:06:21.857] jjoshua2: it wouldnt req2uire the whole compute power of from scratch
3534[14:06:39.341] jjoshua2: there is a lot of NN research in transfer learning and it is very powerful now
3535[14:06:48.037] alexey305472: if anyone could write instructions for distributed Cereballum (i. e brain-engine), perhaps I could run it on my servers.
3536[14:07:23.116] alexey305472: I wonder if Allie is stronger than Leela at this point.
3537[14:07:28.969] jjoshua2: cerebellum is closed source
3538[14:07:38.748] jjoshua2: only the result is released by that author
3539[14:07:49.333] jjoshua2: and the stockfish change that lets it use that opening book
3540[14:08:23.550] alexey305472: if anyone could write an open source version of Cereballum I would commit several servers to work full time on it.
3541[14:08:41.099] talkshowwcharles: Much more compute power would allow for much more experimentation and faster code improvement + more training in general
3542[14:08:42.100] jjoshua2: i open source one on github but it doesnt support more than one machine
3543[14:08:48.052] atanh: @Twipply Because EGTBs are not part of the rules of chess. If you think they are, then ask yourself this: Are human players allowed to consult EGTBs during a tournament chess game? After all, they're 'just part of the chess rules', right?
3544[14:09:02.344] jjoshua2: noobpawn already has one that is working on lots of machines
3545[14:09:05.255] jjoshua2: see !db
3546[14:09:19.666] jjoshua2: they are in ICCF @atanh
3547[14:09:20.545] sometimesithurts: @alexey305472 !chessdbcn
3548[14:09:25.389] Twipply: So what if they aren't a part of the rules?
3549[14:09:26.432] alexey305472: !db
3550[14:09:28.197] eval_bot (37b) +0=0-0 • No moves • No games • <Lichess>
3551[14:09:30.868] Twipply: Surely zero is about zero human knowledge
3552[14:09:33.720] jjoshua2: !chessdbcn
3553[14:09:34.076] Nightbot: A self-learning opening database run by noobpwnftw now with more than 3 billion unique positions. Nf6 -150 stands for 1.5 pawn advantage for the side to move for Nf6. !/*/? is a note of the move(good/neutral/bad), (aa-bb) means there are aa known reply moves of which bb is good+neutral. See also !probe
3554[14:09:37.095] Twipply: this doesn't mean only the rules are allowed
3555[14:10:21.427] Twipply: Unless you think it means uhh, the rules and zero more?
3556[14:10:37.333] talkshowwcharles: TBs are simply truths / facts about positions.
3557[14:11:11.347] sometimesithurts: @alexey305472 chessdbcn is what you are looking for... talk to noob about it, it's getting better and better see !probe for gui
3558[14:11:57.822] tinduz_r: !cdb
3559[14:11:59.127] chessdbcn (38w) • No moves, position queued for analysis. <ChessDBCN>
3560[14:11:59.442] sometimesithurts: !probe
3561[14:11:59.679] Nightbot: To probe !chessdbcn - database and interface created by noobpwnftw: 1. Official: https://www.chessdb.cn/queryc_en/ 2. Ferdy: https://fsmosca.github.io/ChessDB-Online-Book/
3562[14:12:07.000] Twipply: @talkshowwcharles that would be my line of thought, too
3563[14:12:09.218] Vizvezdenec: http://tests.stockfishchess.org/tests/view/5d7d14680ebc5902d3866196 Ka[[a
3564[14:12:11.128] Vizvezdenec: Kappa
3565[14:12:20.150] Twipply: They're truths about the game and thus not human knowledge
3566[14:12:58.964] alexey305472: noobpwn is here?
3567[14:13:17.661] sometimesithurts: not now
3568[14:14:10.069] muppetmuppet: he was earlier today
3569[14:16:23.486] tinduz_r: With how close the Top 3 are right now, I hope we get a bonus of Top 2 vs Top 3 in some form (barring crashes) regardless of sufi.
3570[14:18:23.801] muppetmuppet: maybe even make it top 4 for an even number
3571[14:19:20.122] atanh: @jjoshua2 But not in 'standard' chess as we know it. And even with ICCF, I would wager that the rules let you *consult* an EGTB (or probably any outside source), but that EGTBs *themselves* are not part of the rules. Otherwise, the rules would be many mega/gigabytes in size. EGTBs are *derived* from the rules, but they are not rules themselves. You don't *have* to play a TB winning/drawing move.
3572[14:19:58.118] Twipply: Still wondering why we care that EGTB aren't a part of the rules
3573[14:20:08.787] jjoshua2: @atanh they are actually part of the adjudication rules
3574[14:20:13.462] muppetmuppet: well you don't have to make a legal move either
3575[14:20:18.723] jjoshua2: either player can claim a TB win without playing
3576[14:20:35.723] jjoshua2: so to implement the entire ICCF rules is impossible without much size
3577[14:20:52.685] talkshowwcharles: But if you don't play TB move you play suboptimally. You can never out learn perfection
3578[14:20:57.732] atanh: @Twipply The rules of chess themselves are human knowledge, so it's not zero human knowledge. It's zero human knowledge *beyond* the rules of the game (whether it be chess, Go ,Shogi, or whatever).
3579[14:21:43.785] tinduz_r: I see it more as beyond the truths of the game.
3580[14:22:12.962] muppetmuppet: yes like maths 1+1 =2 is a human zero knowledge truth
3581[14:22:19.019] tinduz_r: TBs are part of the truths of the game. Adding a line that says score more if we have more materials is what's non zero.
3582[14:22:25.905] talkshowwcharles: Me too. No point replacing truths
3583[14:23:18.946] lrmead: k-d2-c2; then b3
3584[14:23:19.905] muppetmuppet: I don't see lczero having to prove all of maths just so it can use multiplication
3585[14:23:51.246] muppetmuppet: I've no idea if it does use multiplication obviously
3586[14:24:34.401] alexey305472: Stockfish will completely own division 2, if it gets degraded. Checkmate.
3587[14:24:57.435] sometimesithurts: div 1 you mean
3588[14:26:01.351] sometimesithurts: league 1 actually
3589[14:26:04.246] muppetmuppet: well time for some bridge so you are free of me
3590[14:26:30.246] tinduz_r: Competitive bridge or casual?
3591[14:26:38.709] muppetmuppet: competitive
3592[14:27:01.499] tinduz_r: Nice. Best of luck.
3593[14:27:02.437] atanh: @talkshowwcharles In the same way that theorems follow from axioms, EGTBs follow from the rules. But they are not part of the rules themselves. They have to be derived. Otherwise, why do we have to store them in huge DBs and why can't we just recall them from our memory of the rules? ...
3594[14:27:12.999] muppetmuppet: I would claim to be about I.M. level at bridge
3595[14:27:32.765] atanh: ... If EGTBs are 'part of the rules', then a 32 pc TB is technically part of the rules. So why can't engines access a 32 pc TB to play the 'correct' move? Because we don’t actually *know* what said 32TB would say is the correct move, because such a TB has not yet been derived. ...
3596[14:28:03.970] atanh: ... We are *limited* in our *knowledge* of the outcome of most moves. We have *gained* such knowledge up to 7pc TBs so far, and that knowledge was derived from *humans*, not from Lc0 learning it itself.
3597[14:28:05.126] muppetmuppet: if it had been chess would be solved
3598[14:28:53.522] muppetmuppet: not that part ofchess is solved once you get to 7 man
3599[14:29:30.779] tinduz_r: Any news on 8-men TB?
3600[14:29:35.975] jjoshua2: @atanh 7 man tb is specifically part of the rules not 8+
3601[14:29:40.402] muppetmuppet: we don't play on with q vs k we just claim a win
3602[14:29:42.322] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
3603[14:29:54.282] jjoshua2: iccf mentions 7 man tb specifically no other number
3604[14:30:01.152] talkshowwcharles: Still reaching tb is win according to rules in this type of event
3605[14:30:09.477] jjoshua2: once 8 man is generated iccf probably will change its rules for iccf chess
3606[14:30:54.081] jjoshua2: in 2016 iccf only allowed up to 6 man tb it appears
3607[14:30:54.145] atanh: If EGTBs are 'just facts about positions', then I guess opening books should be allowed, too, right? I mean, it's just facts about which games have been played historically, and which of those were won/drawn/lost. Just because it's a truth or a fact doesn't mean it's not also *knowledge* that humans have gathered and organized themselves. Lots of facts and truths remain unknown.
3608[14:31:26.449] jjoshua2: i might have heard wrong that iccf changed its rules from 6 to 7
3609[14:31:57.051] talkshowwcharles: Opening play statistics are just that.
3610[14:33:27.613] tinduz_r: The path from opening->w/l/d is different from endgame_tb->w/l/d
3611[14:33:29.010] talkshowwcharles: Might be totally wrong
3612[14:34:25.552] tinduz_r: The type of fact you get from the two is different. One is absolute, the other statistical.
3613[14:34:26.022] atanh: @jjoshua2 Okay, I stand corrected on that part about ICCF. thx for that correction. :) However, I would argue that that is a particular style of chess with special rules that don't apply to the game that most people play when they play standard 'chess'. Also, this objection isn't only about Chess, it should and does also apply to any other game. ... cont'd
3614[14:36:02.380] kanchess: hey jjoshua2 congratz with current #1 place <3
3615[14:36:15.407] kanchess: i hope alliestein will qualify for SuFi
3616[14:37:14.211] atanh: continuation: In a brand new game, just invented, suppose the possibility of using an EGTB is there, but none have been developed yet. And we want an AI that will play this game, given only the rules of that game. Well, since there's no TBs yet, and they aren't specified by the rules, obviously we can't give them to the AI....
3617[14:37:19.549] kanchess: if tcecSF manages to survive this ordeal it may make it there as well
3618[14:37:38.467] kanchess: !news on AI
3619[14:37:38.678] Nightbot: “An AI-equipped surveillance camera would be not a mere recording device, but could be made into something closer to an automated police officer.” --Edward Snowden - see https://bit.ly/2klhCuA
3620[14:39:15.378] atanh: ... But if the AI *itself* could learn and generate such a TB on its own, that would be completely within the 'zero' principle. That is what I argued from the beginning. If Lc0 is going to use any sort of TB, it should be something that the software has derived on its own. ...
3621[14:39:21.015] sometimesithurts: the whole point about the zero thing, is that every knowledge we humans have about a game, might be wrong (or at least not optimal).
3622[14:40:01.944] sometimesithurts: so we let the machine to learn for himself, nad not to use our flawed knowledge.. but Tbs are not a flawed knowledge
3623[14:40:19.971] talkshowwcharles: Eventually you could learn endgame but it takes lots of resources from learning good midgame
3624[14:41:06.779] atanh: ...Besides: Who knows? Maybe by having a general system for generating TBs for any given game, say chess, we might discover better ways of organizing them and using them. For example, currently Leela 'trolls' because she hasn't learned for herself how to use human-generated TBs. If she developed her own TB, maybe it would include info for her to use it more intelligently/efficiently.
3625[14:41:39.589] sometimesithurts: I think this is all about the zero approach... I really don't care if the only input is actually zero or not, but if it could be possibly flawed or not
3626[14:42:05.832] Fabichreportedich: After all we just want the best playing chess program. How it is achieved doesn't really matter right?
3627[14:42:26.566] kanchess: no casualties involved please
3628[14:42:38.251] Fabichreportedich: monkaGiga
3629[14:42:56.615] lmabacus: Acceptable losses
3630[14:43:58.890] tinduz_r: The "spirit" of zero is more in line with not putting assumed truths. The zero approach might be more of gray area if you're saying original rules only.
3631[14:44:00.373] talkshowwcharles: TB often gives many roads to win , leela just often selects among the longer ones
3632[14:44:25.132] kanchess: when will Elsie AsianGlow see the tcecDD and stop this tcecSH
3633[14:44:32.820] AfuroZamurai: Leela dragging out the draw again...
3634[14:44:55.039] atanh: @muppetmuppet Actually, to prove 1+1=2 directly from axioms is very difficult. I doubt more than one or two people here could do it (certainly not me). See us metamath org/mpegif/2p2e4 html (replace spaces with dots) for a proof of 2+2=4. Try drilling down on each of the theorems used in the proof to see how they are proved as well.
3635[14:44:55.763] Vizvezdenec: https://github.com/official-stockfish/Stockfish/pull/2304
3636[14:44:57.798] Vizvezdenec: LUL
3637[14:45:38.785] Stephane_Nicolet: if SF gets downgraded, would it be possible that it re-enters the tournament in Div5, please?
3638[14:46:24.717] Hanamuke: LUL
3639[14:46:35.658] EmteeGaming: Just create a smurf, clone stockfish and submit under a new name
3640[14:47:11.545] fire2fire: Give it the name thisIsNotSFforSure
3641[14:47:20.411] zlatohrbitek: @atanh It's certainly difficult in Metamath's theory, say, or in Russell and Whitehead's theory. But there are also some theories, such as Martin-Lof type theory, in which it's trivial to prove 1+1 == 2. :)
3642[14:47:28.220] twindlers: @atan "But if the AI *itself* could learn and generate such a TB on its own, that would be completely within the 'zero' principle" ... you might as well argue that the AI must self-generate it's own engine code in order to be truly "non-zero" ... since stupid humans had to encode the bytes that are the Leela engine, then it the entire premise of "zero" is already compromised
3643[14:47:29.848] Hanamuke: I think those crashes were a publicity stunt from SF and tcec TheIlluminati
3644[14:47:35.424] gideon425: I was going to create a version of SF that crashes a lot
3645[14:47:42.124] tinduz_r: BrinedFish, since Stockfish is supposed to be unsalted.
3646[14:48:27.652] atanh: @muppetmuppet The point of AlphaZero (and indirectly, Lc0) was not to build up math from scratch. There are other projects doing that, such as the aforementioned MetaMath. A0 was to be able to play certain kinds of *games*. ...
3647[14:48:33.668] gideon425: but as usual the SF team was way ahead of me
3648[14:48:38.381] Hanamuke: 2=1+1 is the definition of 2 in most theories I would say.
3649[14:48:38.547] Vizvezdenec: you will probably hate me for how this PR is done @Stephane_Nicolet Kappa
3650[14:48:58.464] atanh: ...The A0 papers set out the pre-conditions for such games; I won't belabor them here. But they restricted themselves to only supplying the rules of the games, not human-derived knowledge (such as theorems/TBs).
3651[14:49:02.445] Twipply: I was totes afk a whole bunch now
3652[14:49:08.452] Hanamuke: it wwould be in Paeano or ZFC for instance
3653[14:49:17.025] Twipply: I'd sooner say that opening books are not facts of the game, only the thoughts of its players
3654[14:49:33.554] Twipply: or, if you want to go the opening database route, then facts about the players of the game
3655[14:49:41.424] Twipply: neither is knowledge about the game itself
3656[14:49:43.342] twindlers: @atanh TBs are derived by only knowing the rules of the game
3657[14:50:00.195] Vizvezdenec: I think I'm a retаrd and could've made initiative ( Score ) NotLikeThis
3658[14:50:44.204] twindlers: @Twipply totally agree that opening books are not derived from the rules of the game ... but end game TBs are definitely
3659[14:50:55.676] Twipply: Sureeeeeeeeee
3660[14:51:00.589] Twipply: God damn this keyboard
3661[14:51:30.530] tinduz_r: You should differentiate the spirit of zero vs the puristic zero approach. A0 chose a zero approach to try to see the limits of the zero spirit (not putting assumed truths).
3662[14:52:00.234] atanh: @muppetmuppet There's actually a good example. You said " we don't play on with q vs k we just claim a win". Actually, that's not true. The rule is you can resign your position, or your opponent can suggest it. But technically, if you don't resign, the oppo has to checkmate you. The could accidentally stalemate you. ...
3663[14:53:01.881] atanh: ... And also there *are* specific rules about ending games with 'insufficient materials', and these situations are given. That's about the extent of 'tablebase knowledge' that's actually encoded in the rules.
3664[14:53:04.504] Vizvezdenec: so A0 broke it own non-zero approach
3665[14:53:09.863] Vizvezdenec: by adjudicating wins pre-mate
3666[14:53:20.012] twindlers: cheaters!
3667[14:53:28.151] Vizvezdenec: because resigning and adjudicating is what humans bring into chess
3668[14:53:48.626] lrmead: Someone tell me: Why not Kc2 and b3 creating a passed a -pawn ...
3669[14:54:03.433] gideon425: is KNN insufficient?
3670[14:54:13.816] twindlers: sometimes I resign games because my wife tells me dinner is ready ... and if I don't come right away she gets pissed ;-)
3671[14:54:23.302] lrmead: Black can do nothing against this ...
3672[14:54:24.291] tinduz_r: Throughout the series of papers, you could say that. That's why it's differentiated.
3673[14:54:35.182] Twipply: You know, some of these AlphaGo/Zero papers are so garbage I wouldn't even be surprised if they hadn't thought of EGTB
3674[14:54:40.537] sometimesithurts: but wait isn't chess made by humans? shouldn't lc0 itself generate the chess game? LUL
3675[14:54:52.661] Vizvezdenec: it loses d pawn I guess?
3676[14:54:52.924] gideon425: gasp
3677[14:54:57.685] Archstro: Chess was so great before humans brought all these rules and contaminated the game.
3678[14:55:18.481] atanh: @jjoshua2 Just curious (it doesn't affect my argument/position): Does ICCF specify exactly which TBs are allowed to be used? For example, say my cousin has his own special 7pc TB and I use it to show that I'm in a winning position. But oppo thinks my cousin is scamming. Or maybe the TB just has some errors in it. What happens then?
3679[14:55:55.017] Stephane_Nicolet: @Vizvezdenec yes, I shall rewrite it to try to improve the style, but it is always better to put in the PR the exact version which you tested, please (because reformating can be a slowdown and we have to measure it)
3680[14:56:04.156] talkshowwcharles: They play on.
3681[14:56:14.237] jjoshua2: im not sure actually @atanh i think iccf maintains its own copy as the master reference
3682[14:56:16.113] Vizvezdenec: I didn't reformat
3683[14:56:16.451] Stephane_Nicolet: for instance the std::max() :-)
3684[14:56:19.354] tinduz_r: I'm more into the "let's see what the machine can learn about the game if we don't assume what's best" part, rather than "let's see if the machine can be the best using only the basic rules of chess".
3685[14:56:26.287] Vizvezdenec: ah
3686[14:56:31.561] Twipply: Their paper does actually say "without any additional domain knowledge except the rules of the game" which would exclude EGTB
3687[14:56:36.754] Vizvezdenec: well is std::max slower than if?
3688[14:56:40.582] Twipply: but then they also say it's "tabula rasa", which wouldn't exclude EGTB
3689[14:56:59.657] Vizvezdenec: sorry I'm not entirely sober and thought it will be better (at least shorter for sure)
3690[14:57:00.902] jjoshua2: they also say EGTB is for future research
3691[14:57:03.541] Vizvezdenec: maybe I'm not right Kappa
3692[14:57:17.120] Stephane_Nicolet: we don't know, but it is better to tell in the PR, IMHO
3693[14:57:24.617] Twipply: They're dumbasses for not using EGTB anyway
3694[14:57:38.056] twindlers: @tinduz_r can you give an example of where those two philosophies might lead to different outcomes or decisions?
3695[14:58:13.692] Vizvezdenec: I mean I basically do std::max instead of (value > 0)
3696[14:58:18.216] jjoshua2: it wouldnt play on @talkshowwcharles the whole purpose of the rule is to prevent someone from tyring to stall to prevent a loss by taking makimum amount of time possible to delay loss or even draw i suppose
3697[14:58:23.448] Vizvezdenec: I think we do it everywhere if we can (?)
3698[14:58:34.201] jjoshua2: at 3 days / move a 200 move endgame takes 600 days if u drag it out
3699[14:58:44.866] Vizvezdenec: instead of (if (value > 0)) geez
3700[14:58:58.247] tinduz_r: An example would be what we're discussing right now. Endgame TBs are not human assumptions, they're truths. The puristic "only basic rules" disallow TBs then, but the zero spirit of "no assumed truths" allow it.
3701[14:59:19.101] kanchess: this happened toDM jjoshua2 just before he became WC iirc
3702[14:59:32.496] kanchess: some guy dragging out the loss over max time allowed
3703[14:59:40.694] kanchess: like months
3704[14:59:45.976] Nightbot: We need your help and support: TCEC is happening thanks to the support of our community. If you too want to help TCEC, and help us grow and develop, please support by cheering or by using the buttons to donate or subscribe below our chat at https://tcec-chess.com . Every little bit helps and is very much appreciated! nadedKing
3705[15:00:01.778] talkshowwcharles: If you claim to have a 9pc tb that would not count. Have to be a publicly proven tb
3706[15:00:12.200] Vizvezdenec: I mean we can essentially substitute every std::max
3707[15:00:12.799] djcat85: the point of "zero" was no domain-specific knowledge (beyond the rules), so that the NN learns "everything"
3708[15:00:28.653] Vizvezdenec: with (value1 > value2) ? value1 : value2
3709[15:00:32.890] Vizvezdenec: but we don't do it
3710[15:00:42.449] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
3711[15:00:44.206] atanh: @talkshowwcharles No, it's a fact that So-and-so beat Whats-his-name in such-and-such position in 1997 somewhere. And the same with the billions of other facts in a good opening book. And counting the total wins/draws/losses from the same pos, well that's just another truth from those other facts, so clearly it's not human knowledge if it's just an innocent little bunch of chess facts, right?
3712[15:00:47.637] djcat85: Maybe train an NN to adjudicate games :)
3713[15:00:52.276] Vizvezdenec: so it's okay to rewrite if (value > 50) into std::max(value - 50, 0)
3714[15:01:04.006] twindlers: @tinduz_r I disagree that "only basic rules" disallows TBs - but I can see how some purists could take that stance since TBs are derived from basic rules - but are not themselves the rules
3715[15:01:55.896] Stephane_Nicolet: do we have any news of DragonMyst ?
3716[15:02:08.761] tinduz_r: @twindlers I agree with you to be honest. I'm not too puristic on the "only basic rules", since ICCF rules even allow TBs. But I'm willing to concede that point since that was not my point anyway.
3717[15:02:15.462] djcat85: @twindlers think of the generalizability to other games----basically the same "zero" algorithm applies to Go, Shogi, checkers, etc. If it were trained with TBs, it's already much more chess specific.
3718[15:02:18.703] kanchess: no except that he is absent for reasons of health Stephane_Nicolet
3719[15:02:28.178] Stephane_Nicolet: aie
3720[15:02:37.729] Vizvezdenec: he makes sf regression tests infographics
3721[15:02:55.496] Vizvezdenec: I kinda was not motivated to do them when sf had +16 elo regression tests for 2 months
3722[15:02:58.372] mannimond: alli scares me
3723[15:03:05.835] Vizvezdenec: and he picked them up
3724[15:03:07.137] Vizvezdenec: again
3725[15:03:41.195] atanh: @tinduz_r No, recording known historical events about chess games in a database is just as factual as recording the historical events about EGTB generators having been into a database. How do you know the EGTB doesn't have errors in it? You don't. Because it's compiled human knowledge.
3726[15:03:51.981] talkshowwcharles: But that doesn't say anything about the game. There are no known perfecly played games of chess
3727[15:03:53.060] lmabacus: @kanchess Was it the game at the end of this article? https://en.chessbase.com/post/better-than-an-engine-leonardo-ljubicic-1-2
3728[15:04:02.202] kanchess: lemme look
3729[15:04:03.041] sometimesithurts: you clould just simply #define max(a,b) a>b?a:b at the top cuz it's faster.. I used to do that in ICPC-ACM :)
3730[15:04:09.225] tinduz_r: @atanh It's brute-force proven.
3731[15:04:41.090] tinduz_r: @atanh If we're talking math, proof by brute-force.
3732[15:05:04.137] kanchess: yes lmabacus you can see in the annotations the game unnecessatily dragged on for 1 YEAR
3733[15:05:16.178] kanchess: 1 YEAR yes
3734[15:05:24.749] twindlers: @djcat85 I'm not sure I understand ... are you saying Leela can play go by being trained on chess games ... but since the training uses EGTBs then that's why Leela can't play go very well?
3735[15:05:27.629] kanchess: just to stall the loss
3736[15:05:34.682] lmabacus: Yeah, when I first read that I thought he was just joking. Geez, a whole year.
3737[15:05:48.781] kanchess: un-gentleman-like
3738[15:05:52.654] Vizvezdenec: https://csgostats.gg/player/76561198288223041#/matches look at my latest matches PogChamp
3739[15:05:57.828] twindlers: I don't see how an engine could be trained to learn chess and then be good at go without being trained on go itself?
3740[15:06:01.581] kanchess: un-sportsman-like too
3741[15:06:12.248] lmabacus: Definitely
3742[15:06:28.185] jim_vs: what are you talking about? @kanchess
3743[15:06:37.989] kanchess: see link posted by lmabacus
3744[15:06:49.670] kanchess: and annotations about durations in the game at the end
3745[15:06:51.807] jim_vs: I cannot. Came to late
3746[15:06:55.751] djcat85: @twindlers No, I'm saying the same "zero" learning algorithm can learn chess, Go, Shogi, etc. OTOH, a learning algorithm fed TBs, is chess specific.
3747[15:06:58.705] jim_vs: too*
3748[15:07:00.648] kanchess: https://en.chessbase.com/post/better-than-an-engine-leonardo-ljubicic-1-2
3749[15:07:03.731] lmabacus: This link: https://en.chessbase.com/post/better-than-an-engine-leonardo-ljubicic-1-2
3750[15:07:07.287] jim_vs: thx
3751[15:07:40.544] kanchess: interview with Dragon Mist from this chat
3752[15:07:41.561] Twipply: @djcat85 so make the algorithm "use EGTB if they exist" and now it's not chess specific anymore?
3753[15:07:51.129] Twipply: Though EGTB can exist in a variety of games I'm sure, not just chess
3754[15:07:53.526] twindlers: @djcat85 wouldn't a go program be fed go EGTBs?
3755[15:08:02.714] Twipply: Go EGTB don't exist
3756[15:08:04.816] tinduz_r: On a side note, Leela eval graph is more erratic than RF this game. Probably because it's the one playing?
3757[15:08:10.552] Occyroexanthub: others(someone that he did beat Deep Thought in the 80s) have done this bad behavior also dragging a lost game for years :P
3758[15:08:14.218] Twipply: but yes, clearly
3759[15:08:35.672] kanchess: tcecLC has a big blind spot for tcecDD
3760[15:08:54.954] saurus1969: LEELA is getting optimistic LUL
3761[15:09:14.983] tinduz_r: Time troubles abound.
3762[15:09:24.302] kanchess: yes
3763[15:09:36.702] kanchess: ahead*
3764[15:09:53.443] atanh: @zlatohrbitek The point is a general one, not specific to MetaMath or Chess or Go or any particular axiomatic system. The point is simply that axioms are given, and theorems are derived (and not given). The zero approach gives certain axioms (the NN and training architecture, and the rules of a particular game, say Chess), and the point is to see how far it can get just from that info alone.
3765[15:10:12.384] djcat85: @twindlers I'm not a Go player, but my very non-expert understanding is that Go becomes more complex as the game goes on, so TBs aren't useful. At any case, they are not "zero".
3766[15:10:15.617] twindlers: @atanh arguing that EGTBs aren't derived from chess rules because they might have bugs is like arguing that Leela engine isn't learning chess because the Leela engine might have bugs
3767[15:10:17.997] kanchess: sac PogChamp
3768[15:10:24.533] Archstro: Is there any statistical proof that gentleman-like behavior leads to more success in life?
3769[15:10:34.720] halberder: evaluating this tournament by Instant Runoff (repeatedly eliminate the lowest scoring engine and its games), currently Leela and Allie are tied for first, followed in order by [houdini = stockfish] komodo, stoofvlees, komodoMCTS, and scorpionn
3770[15:10:38.148] kanchess: yes Archstro
3771[15:10:44.117] saurus1969: Ok, draw soon
3772[15:10:45.298] Hanamuke: @atanh if rules are the theory, egtb are theorems
3773[15:11:17.533] tinduz_r: ^
3774[15:11:31.393] kanchess: in sociobiology it is stated that a single antisocial person has a good chance of survival in a group of social persons, but a society of antisocial persons will never be as successful as a society of social persons
3775[15:11:34.211] twindlers: @djcat85 egtbs are "zero" if they are derived by only knowing the rules of the game.
3776[15:12:08.807] twindlers: I don't play go either, so I'm not sure if egtbs would be useful in go or not ... but just because they are useful in chess and not in go doesn't make them non-zero
3777[15:12:43.919] atanh: @twindlers "the AI must self-generate it's own engine code". No, that's part of the 'givens'. The engine, sans rules-of-the-game is given. You're then to put together a NN scheme that can represent the game, and supply the game rules, and then flip the switch. There's nothing about bootstrapping the whole engine. The constraints are in the A0 papers.
3778[15:12:54.743] halberder: @kanchess that depends on how effectively the "rule abiders" punish the "rule breakers" (social vs antisocial are not the right words; someone can be rule abiding but not social)
3779[15:13:08.781] tinduz_r: Boom? Fake?
3780[15:13:09.732] djcat85: @twindlers they are not "zero" from the point of view of wanting a *domain-general* learning algorithm that can learn to play almost any game. Many chess fans struggle with this point, because they're imagination is too limited to be able to imagine other goals in life than the best possible chess.
3781[15:13:22.269] Archstro: It seems to me that most of the powerful people in the world did not get to where they are by abiding by moral guidelines that we're taught to follow.
3782[15:13:23.834] Hanamuke: howw would you do egtb for go, you have 2^121 possible ending position
3783[15:13:25.051] kanchess: well, sociobiology speaks of "individuals" and of "cheating" behavior to be more precise
3784[15:13:35.092] kanchess: i rephrased it a bit
3785[15:13:40.325] saurus1969: Wow tho might be interesting
3786[15:13:42.092] twindlers: @atanh "the engine code might have bugs though, and might encode the rules of the game incorrectly ... so it can't be zero because it is built by flawed human hands." Isn't that the same as your argument against egtbs?
3787[15:13:53.054] theo77186Fr: I just read the discussion about std::max and stuff and I did some experiments with my compiler.
3788[15:13:56.431] LeeTCEC: But the individual is just the genes way of making more genes.
3789[15:14:01.181] halberder: yes, "cheaters" vs "non-cheaters" are also good words for the situation. The success of the cheater depends on how well the non-cheaters can identify and punish the cheaters
3790[15:14:10.089] atanh: @twindlers "TBs are derived by only knowing the rules of the game" Keyword: Derived. Critical question: Derived by whom?
3791[15:14:18.368] theo77186Fr: Strangely using 3 different codes to do the same thing generates different code
3792[15:14:24.952] twindlers: @djcat85 but if a domain does have useful egtbs then it is a general purpose tool, right?
3793[15:14:38.387] twindlers: @atanh by engine code
3794[15:15:13.581] kanchess: success varies acc to observations in different species. but the rule remains that a society of cheaters is always less successful
3795[15:15:16.918] Archstro: Just the fact that women find bad boys more attractive than nice guys could be proof that being bad is an evolutionary advantage.
3796[15:15:17.972] Dax_Maclaine: I just got here what is alliestein?
3797[15:15:41.105] kanchess: !tcec
3798[15:15:41.292] Nightbot: Top Chess Engine Championship formerly known as Thoresen Chess Engines Competition (TCEC or nTCEC) is a computer chess tournament that was organized, directed, and hosted by Martin Thoresen and now organized by Chessdom. It is often regarded as the Unofficial World Computer Chess Championship because of its strong participant line-up and long time control matches on high-end hardware, giving rise to very high-class chess. See also https://goo.gl/zuTijZ
3799[15:15:44.986] kanchess: welcome here
3800[15:15:53.986] trajkoskigoran02: @Dax_Maclaine NN
3801[15:16:00.421] marklefler: How can these nn programs play so well and not know this is a draw?
3802[15:16:05.617] Ove__: !alliestein
3803[15:16:05.926] Nightbot: See !allie and !leelenstein
3804[15:16:08.368] Vizvezdenec: @Stephane_Nicolet I think it was your patch to use fraction of kingdanger in endgame eval? PogChamp
3805[15:16:12.286] djcat85: @twindlers no, some games might allow for something analogous to TBs, others don't, so by relying on TBs you already restrict yourself. For example, your TB-reliant learning algorithm won't be applicable to Go.
3806[15:16:21.322] atanh: @Twipply The records in opening books are merely records of games of chess. These (in an accurate opening book) are just facts about chess. The games themselves are easily derived from the rules of chess, as you can see for yourself by looking at any of the games.
3807[15:16:24.740] sparkleofdiego: @marklefler you tell us, you are the programmer! LUL
3808[15:16:29.024] Ove__: !allie
3809[15:16:29.194] Nightbot: Allie is a neural network based engine written by Adam Treat, inspired by the AlphaZero paper and the Lc0 project. See - https://github.com/manyoso/allie , https://www.patreon.com/gonzochess75 and http://www.chessdom.com/alliestein-the-new-neural-network-entering-tcec-s15/ , Join the discord channel to follow Allie - https://discord.gg/UnbsJRU
3810[15:16:32.237] Vizvezdenec: now we will have fraction of compleity used in midgame eval
3811[15:16:35.523] lunarzale: @Dax_Maclaine It's a mix of NN and traditional alpha/beta search
3812[15:16:42.445] Stephane_Nicolet: yes
3813[15:16:42.847] Hanamuke: you don't need to recogniz drawn endgame if you can recognize non-draw endgames
3814[15:16:47.075] Vizvezdenec: I guess the last thing standing will be scalefactor Kappa
3815[15:16:56.242] Dax_Maclaine: k ty I never heard of it b4
3816[15:16:57.163] Vizvezdenec: midgame scalefactor someone? Kappa
3817[15:17:14.074] twindlers: @djcat85 why do you have to rely on egtbs? someone already said, your general purpose algorithm would be to use egtbs if they exist, otherwise, don't use them.
3818[15:17:15.755] Stephane_Nicolet: it was my patch to introduce the initiative() idea in SF too :-)
3819[15:17:22.785] tinduz_r: Draw if leela doesn't take.
3820[15:17:30.345] Vizvezdenec: I think we should rename it
3821[15:17:34.838] Vizvezdenec: into smth like winnability
3822[15:17:42.313] Vizvezdenec: I mean... It's not initiative?
3823[15:17:51.463] Vizvezdenec: it more like "how winnable this position is"
3824[15:17:54.209] Hanamuke: LUL winnability
3825[15:18:03.481] Vizvezdenec: complexity is a good word :)
3826[15:18:11.161] kanchess: !define complexity
3827[15:18:11.537] Nightbot: goto https://en.wiktionary.org/wiki/complexity nadedKing
3828[15:18:18.025] Vizvezdenec: much closer in meaning to what this eval part actually does
3829[15:18:22.416] tinduz_r: Or how likely the opponent will make a mistake.
3830[15:18:23.895] atanh: @jjoshua2 If they have their own master copy, then I would argue that they actually do include entire EGTBs in their official rules. Most games of chess don't play by those rules of course.
3831[15:18:32.734] Vizvezdenec: although you like strange names Kappa
3832[15:18:40.215] Vizvezdenec: Cough Cough tropism Cough
3833[15:18:45.103] sometimesithurts: @Vizvezdenec I just checked and it seems both std::max and a>b?a:b generate the exact same binary with compiler optimization on
3834[15:19:06.722] Vizvezdenec: std::max is shorter
3835[15:19:22.558] twindlers: and easier to read
3836[15:19:25.020] Vizvezdenec: you can wrap up if and definition of stuff with 0 into 1 line of code
3837[15:19:31.392] sometimesithurts: yeah but I thought a>b?a:b is faster but it's not
3838[15:19:34.484] atanh: @Twipply Tabula rasa is the definition of excluding TBs! Erase the table!
3839[15:19:47.155] kanchess: this does look like a 800 elo game
3840[15:20:09.999] twindlers: it would be pretty strange to have a std:: algorithm that is slower than an alternative
3841[15:20:35.319] kanchess: "see i know how the pieves move"
3842[15:20:37.918] twindlers: @atanh LUL
3843[15:20:41.354] kanchess: pieces*
3844[15:20:53.061] Twipply: Mebbi
3845[15:21:02.541] twindlers: I know the pieces fit!
3846[15:21:06.747] kanchess: :D
3847[15:21:10.525] theo77186Fr: std::min and std::max got inlined anyway
3848[15:21:26.653] tinduz_r: Draw.
3849[15:21:41.396] kanchess: phew
3850[15:21:48.178] theo77186Fr: But I find strange that using std::max generates different code than writing manually
3851[15:21:56.748] tinduz_r: SF and Leela still tied.
3852[15:22:11.506] kanchess: and next, the most exciting game of the day?
3853[15:22:21.420] kanchess: stoof vs stein
3854[15:22:47.126] LeeTCEC: any game where SF doesn't crash?
3855[15:22:50.072] sometimesithurts: @theo77186Fr did you compiled with optimization on?
3856[15:23:05.247] theo77186Fr: Even more shocking: std::max(x+50,0) doesn't generate branches in rv64 but writing manually does
3857[15:23:11.803] theo77186Fr: @sometimesithurts with -O3
3858[15:23:26.645] gnureligion: Ah, now for the "real deal" AllieStein.
3859[15:23:27.096] theo77186Fr: weirdly all cases generates branches in x86-64
3860[15:23:38.808] Aloril42: (script) 94: 2. tcecLC 3836(-2) - 4. Komodo 3809(+2) tcecDD https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-94-lczero-v0-22-0-nt40b-4-160-komodo-2381-00/
3861[15:23:47.495] theo77186Fr: despite branchless versions exist
3862[15:23:56.502] kanchess: !db
3863[15:23:57.807] eval_bot (9w) +2=11-1 • Bb2 (+0=8-1) • e3 (+2=1-0) • cxd5 (+0=1-0) • [1992] Polugaevsky, Lev (2630) vs Kasparov, Garry (2770): draw • [1993] Nikolic, Predrag (2625) vs Dolmatov, Sergey (2630): draw • <Lichess>
3864[15:24:01.435] kanchess: !reverse
3865[15:24:01.596] Moobot: ʇoq‾lɐʌǝ ǝsɐǝld ǝsɹǝʌǝɹ¡
3866[15:24:04.958] eval_bot Game #67 "AllieStein v0.5-dev_1359f44-n10" vs. "Stoofvlees II a12": [book] 9. e3 a4 10. Ba3 Re8 11. Qe2 Bf5 12. Rfc1 e6 13. b4 Nbd7 14. b5 Qa5 15. Nh4 Bg4 16. f3 Bh5 17. bxc6 bxc6 18. g4 e5 19. dxe5 Nxe5 20. gxh5 d4 21. hxg6 hxg6 22. Nf1 dxe3 23. f4 Nd3 ... 1-0
3867[15:24:14.075] atanh: @tinduz_r Even if it's brute force proven, how do you know it has no errors in it? Did you prove it yourself? If not, did you verify the proof yourself? It's remarkably easy to write an incorrect 'proof'.
3868[15:24:14.455] theo77186Fr: Compilers are strange creatures indeed
3869[15:24:15.211] lunarzale: @theo77186Fr it'll generate a branch for a>b?a:b if a is x+50 since it can't optimise away the fact that x might have changed, surely?
3870[15:24:20.781] Occyroexanthub: ah the game where Stoof gifted 2 pieces
3871[15:24:27.842] Aloril42: (script) if 1-0: Sto: 3762(+2), All: 3849(-10); if tcecDD : Sto: 3760, All: 3857(-2); if 0-1: Sto: 3759(-1), All: 3867(+8); simulation and estimated start times: https://tcecbayeselo.chessdom.org/2019/09/14/tcec-season-16-divisionp-game-95-speculation-simulation/
3872[15:24:38.562] Occyroexanthub: for some compensation but was never enough
3873[15:24:54.521] theo77186Fr: @lunarzale x is not declared volatile so it cannot have changed in the middle of nowhere
3874[15:25:17.380] twindlers: @atanh now you are just getting to the point of Descartes ...
3875[15:25:28.629] tinduz_r: @atanh That's stretching pedantry. You could say that for every theorem in math. If it's about making sure, then you could argue it's peer reviewed to some degree.
3876[15:26:08.238] Hanamuke: well, code correctness is a study field, but of course for pratical purpose we only need to test that they give elo
3877[15:28:01.582] kanchess: !db
3878[15:28:02.643] eval_bot (9b) +2=1-0 • a4 (+2=0-0) • Nbd7 (+0=1-0) • [1993] Nikolic, Predrag (2625) vs Dolmatov, Sergey (2630): draw • [2013] Prizant, Jaroslav (2537) vs Ernst, Thomas (2387): white • <Lichess>
3879[15:28:29.734] atanh: @twindlers "arguing that EGTBs aren't derived from chess rules because they might have bugs " I didn't argue that. Please don't put words in my mouth. If you want to make a specific point about something specific I've argued, please quote it directly without paraphrasing. Thanks. :)
3880[15:29:42.592] Nightbot: Reminder for those who support TCEC via Amazon Prime. First link your Amazon account directly from amazon.com to your Twitch account before trying to get the free sub. You must manually renew your sub at https://www.twitch.tv/tcec_chess_tv each month. Thank you! nadedKing
3881[15:31:19.074] atanh: @djcat85 Good points, and well stated also. :)
3882[15:31:41.242] tinduz_r: Anyway. 3:30 AM here. Need to sleep. No crashes please. Good night.
3883[15:31:51.368] kanchess: nn
3884[15:31:51.521] Nightbot: drezzZZZ
3885[15:32:39.725] kanchess: !db
3886[15:32:41.089] eval_bot (10w) +2=0-0 • Ba3 (+2=0-0) • [2013] Prizant, Jaroslav (2537) vs Ernst, Thomas (2387): white • [2017] Rajdeep, Sarkar (2312) vs Szente Varga, F. (2190): white • <Lichess>
3887[15:33:00.495] kanchess: !cdb
3888[15:33:01.698] chessdbcn (10b) • Re8 -12 ! (23-07) • Nbd7 -20 * (20-07) • pv Re8 Re1 Qb6 h3 h6 g4 Nbd7 Qc2 • <ChessDBCN>
3889[15:33:26.566] twindlers: @atanh you wrote something to that effect, but it seems to have scrolled out of the chat window now, your followup to it was "Even if it's brute force proven, how do you know it has no errors in it? Did you prove it yourself? If not, did you verify the proof yourself? It's remarkably easy to write an incorrect 'proof'."
3890[15:34:11.492] twindlers: the post previous to that you stated something to the effect that EGTBs couldn't be counted as derived from chess rules because they might have bugs in them
3891[15:34:31.064] djcat85: @atanh thanks :)
3892[15:34:49.769] twindlers: which then led to a pedantic argument about whether you can believe anything unless you prove it yourself ...
3893[15:35:05.385] atanh: @twindlers ""the engine code might have bugs though, and might encode the rules of the game incorrectly ... so it can't be zero because it is built by flawed human hands." Isn't that the same as your argument against egtbs?" No! And that's not an actual quote of mine, either! NotLikeThis It could be absolutely perfect, but it would still be *human* knowledge, *not* learned from the AI itself. !=0
3894[15:36:00.070] Occyroexanthub: can anyone decipher this 12c5c5 ? Kappa
3895[15:36:15.135] twindlers: @atanh obviously you are carrying on too many threads with too many chatters here to remember even the gist of your own arguments
3896[15:36:18.428] kanchess: not quite @twindlers this is what @atanh said: No, recording known historical events about chess games in a database is just as factual as recording the historical events about EGTB generators having been into a database. How do you know the EGTB doesn't have errors in it? You don't. Because it's compiled human knowledge.
3897[15:37:06.980] twindlers: @kanchess that's not the post I was referring to ... like I said, @atanh is having several threads of conversations with various folks
3898[15:38:00.210] kanchess: mkay
3899[15:38:06.318] twindlers: @atanh was arguing one thread about whether or not egtbs counted as zero or not ... and a main point of the argument against egtbs being zero was that they might have bugs that therefore cause them to not reflect the rules of the game.
3900[15:38:15.700] atanh: @twindlers "[Me]: Critical question: Derived by whom? [You:] by engine code" Not by Lc0's engine code! From Lc0's perspective, this information is handed to her on a silver platter from out of the blue. She didn't learn it herself. Hence, not zero.
3901[15:38:32.366] kanchess: !db
3902[15:38:33.624] eval_bot (12w) +0=0-0 • No moves • No games • <Lichess>
3903[15:38:37.085] kanchess: !cdb
3904[15:38:38.330] chessdbcn (12w) • Rfc1 109 ! (01-00) • pv Rfc1 Ra6 b4 • <ChessDBCN>
3905[15:38:54.555] twindlers: I pointed out that LC0 engine code also has the rules of the game encoded in the LC0 engine ... but since that could also have flaws in the encoding of the rules in the LC0 engine then the LC0 engine itself could be argued to not follow the rules of the game - following @atanh 's arugment against egtbs
3906[15:39:04.876] kanchess: 11. ...Bf5 deviates from reverse btw
3907[15:39:19.864] kanchess: h6*
3908[15:39:25.735] kanchess: reverse had Bf5 i mean